CN110008360A - Vehicle target image data base method for building up comprising specific background image - Google Patents
Vehicle target image data base method for building up comprising specific background image Download PDFInfo
- Publication number
- CN110008360A CN110008360A CN201910278860.4A CN201910278860A CN110008360A CN 110008360 A CN110008360 A CN 110008360A CN 201910278860 A CN201910278860 A CN 201910278860A CN 110008360 A CN110008360 A CN 110008360A
- Authority
- CN
- China
- Prior art keywords
- image
- vehicle
- vehicle target
- specific background
- target image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The present invention includes the technical solution of the vehicle target image data base method for building up of specific background image, it is related to the traffic control system of road vehicle, first collection of material specific background image, production only includes the vehicle target image of single car image afterwards, it again will only include that the vehicle target image of single car is embedded into the vehicle target image that formation in specific background image includes specific background image, it repeats the above steps, prepare the vehicle target image data base comprising specific background image for meeting the needs of deep learning neural network is to image data base and being verified using deep learning neural network, it overcomes in the prior art, it is present under different illumination and weather condition in multiple periods and carries out shooting with video-corder video and Image Acquisition vehicle target image pattern of taking pictures, using the vehicle target figure cut and mirror imageization transformation in left and right is prepared As database, being continuously increased and updating, the defect that manufacturing cost is high and the time is long for different background and vehicle vehicle is not adapted to.
Description
Technical field
Technical solution of the present invention is related to the traffic control system of road vehicle, specifically includes specific background image
Vehicle target image data base method for building up.
Background technique
Identification vehicle target is the core technology in the important directions and intelligent traffic control system of intelligent driving.By
Major progress is achieved in computer vision field in recent years in deep learning neural network, is based on data acquisition, pretreatment, spy
The shallow-layers neural network recognization vehicle targets such as sign extraction, classification are gradually by the deep learning neural network recognization vehicle of high discrimination
Replaced target.The key of deep learning neural network recognization vehicle target is network design configuration to be optimized, but this is necessary
There are a large amount of vehicle target image data bases, is just able to satisfy the demand of trained deep learning neural network.
In order to establish a large amount of vehicle target image data bases to meet the needs of trained deep learning neural network, in road
The traffic control system of vehicle in the prior art, through frequently with having prepared disclosed vehicle target database and video presence
The method for acquiring video.For example, paper (computer and digital engineering, 2019,46 (9), 1871-1875+1915) utilizes Microsoft
COCO public database；Paper (IET Intelligent Transport Systems, 2018,12 (3): 186-193；It calculates
Machine engineering and application, 2018,54 (18): 154-160) use CompCars database；Paper (IEEE Transactions on
2018,27 (1): Image Processing 432-441) uses KITTI and PASCAL VOC2007 database；Paper (IEEE
Transactions on Intelligent Transportation Systems, 2019,20 (2): 749-759；Computer
With modernization, 2017 (8), 264:56-600) research group of author using respective electronics shoots with video-corder the method taken pictures, obtain vehicle
Destination image data library；CN106846813A discloses the method for building urban road vehicle image data base, is using practical
Acquisition is containing Video Quality Metric at the strategy of image；CN109446973A discloses a kind of vehicle based on deep neural network image recognition
Localization method is to acquire vehicle target image sample in multiple periods under different illumination and weather condition；
CN107633220A discloses a kind of vehicle front target identification method based on convolutional neural networks, needs to obtain a large amount of vehicles
Traffic associated picture carries out the transformation of left and right mirror imageization as sample, to the image of collection, and EDS extended data set simultaneously makes label.
The above-mentioned prior art shoots with video-corder vehicle target using existing database and electronics, and video to original acquisition or
Person's image is cut, and true and reliable vehicle target image data base is provided for deep learning neural network, can train depth
Spend learning neural network, but in the presence of planar defect: 1. really acquisition vehicle target image and video limit the kind of database
Class and quantity, it is necessary to which increase shoots with video-corder the time or number of taking pictures, and could improve the capacity of database；2. the side for shooting with video-corder or taking pictures
Method can not can only represent part comprising the vehicle target image in all any scenes, the vehicle target image data base of formation
Vehicle target image under background shoots vehicle target image in specific region and season, cannot be adapted to other regions and season
Vehicle target image under background；3. according to the capacity for increasing video time raising vehicle target database, man power and material
Consumption can increase with the time is shot with video-corder.
Be summed up, road vehicle traffic control system in the prior art, be present in different illumination and day gas bar
The vehicle target image pattern for carrying out shooting with video-corder video and Image Acquisition of taking pictures under part in multiple periods, using cutting and left and right mirror
The prepared vehicle target image data base of pictureization transformation does not adapt to being continuously increased and updating for different background and vehicle vehicle,
And increase the defect of the length of costly and time of vehicle target image data base quantity.
Summary of the invention
It is built the technical problems to be solved by the present invention are: providing the vehicle target image data base comprising specific background image
Cube method, first collection of material specific background image, rear production only includes the vehicle target image of single car image, then will only be wrapped
Vehicle target image containing single car, which is embedded into specific background image, forms the vehicle target figure comprising specific background image
Picture repeats the above steps, and prepares and meets the needs of deep learning neural network is to image data base and using deep learning mind
The vehicle target image data base comprising specific background image verified through network, effectively overcomes in road vehicle
Traffic control system in the prior art, is present in and carries out shooting with video-corder video and bat in multiple periods under different illumination and weather condition
According to Image Acquisition vehicle target image pattern, using the vehicle target image data cut and mirror imageization transformation in left and right is prepared
Library does not adapt to being continuously increased and updating for different background and vehicle vehicle, and increases the expense of vehicle target image data base quantity
With the defect that the high and time is long.
The present invention solves technical solution used by the technical problem: the vehicle target image comprising specific background image
Database building method, the prepared vehicle target image data base comprising specific background image will use deep learning nerve
Network is verified, the specific steps are as follows:
The first step, collection of material specific background image:
By taking pictures or network download method, collection of material specific background image, institute's collection of material specific background image are wanted
Ask as follows:
(1.1) picture material included in specific background image:
Picture material included in specific background image is that carriageway image adds Tree image, building object image, mountain range figure
1~4 kind in picture and road sign image；
(1.2) the pixel size of specific background image:
Specific background image slices vegetarian refreshments size is N pixel × M pixel, i.e., it is laterally M pixel, specific back that longitudinal, which is N pixel,
The vehicle target image slices vegetarian refreshments to be prepared for scape image slices vegetarian refreshments >=included by background image；
(1.3) developed width that specific background image bottom unit pixel indicates is calculated:
The width in established standards lane is L meters, according to the occupied I pixel in lane in specific background image bottom, with such as
Lower formula (1) calculates the developed width R that the bottom unit pixel of specific background image indicates,
In above-mentioned formula (1), the unit of R is rice/pixel；
Second step, production only include the vehicle target image of single car image:
By taking pictures or network download method, acquire and make only include single car image a vehicle target image, should
Only the requirement of the vehicle target image comprising single car image is as follows:
(2.1) direct picture of simple target vehicle is acquired；
(2.2) according to the size and shape of the vehicle of institute's collection of material vehicle target image, it is divided into different automobile types；
(2.3) when take pictures or the vehicle target image of network downloading in comprising corresponding original background image when, pass through figure
It so only include simple target in the vehicle target image for acquiring and converting as processing software deletes the original background image
Vehicle image, the vehicle width in the vehicle target image are J pixel, acquire the vehicle width pixel J in vehicle target image
It is greater than the occupied I pixel in lane in 0.9 times of specific background image bottom；
(2.4) vehicle width corresponding pixel points are calculated:
According to different automobile types described in above-mentioned (2.2) step, measuring its width is W meters, is calculated with following formula (2) corresponding
The vehicle laterally corresponds to the pixel D of the specific background image of above-mentioned first step collection of material,
In formula (2), fix () indicates to be rounded；
Third step, will be in the vehicle target image insertion specific background image only comprising single car image:
The only vehicle target image comprising single car image that above-mentioned second step is made is embedded in above-mentioned first step acquisition
In the specific background image of production, the specific operation method is as follows:
(3.1) size of vehicle target image is adjusted:
The vehicle target image comprising single car image of above-mentioned second step production is calculated separately with following formula (3)
Horizontal and vertical regulation coefficient E,
The size that vehicle target image is adjusted according to the numerical value of regulation coefficient E, the vehicle target image after being sized
For in the vehicle target image insertion specific background image of next step；
It (3.2) will be in the specific background image of the vehicle target image insertion collection of material only comprising single car image:
The carriageway image near center location shown in the specific background image of above-mentioned first step collection of material, in insertion
(3.1) step adjusted size of vehicle target image is stated, which is placed on above-mentioned first step collection of material
Position more than specific background image bottom, the vehicle target image front wheel landing, tire is occupied to be acquired in the above-mentioned first step
Lane position in the specific background image of production, which is opaque；
(3.3) size according to lane in the specific background image of above-mentioned first step collection of material determines the specific background figure
The developed width that image width degree pixel indicates, combining target vehicle width determine that the vehicle target image occupies in specific background image
The quantity of horizontal pixel；
(3.4) adjustment only includes the quantity of the vehicle target image slices vegetarian refreshments of single car image, to keep the vehicle mesh
The laterally and longitudinally same ratio of logo image changes, and only changes the size of the vehicle target image, vehicle shape remains unchanged；
Thus the vehicle target image of the single car image comprising specific background image is made；
4th step, production include the vehicle target image data base of specific background image:
The step of repeating above-mentioned second step to third step prepares comprising specific background image all different automobile types
Single car image vehicle target image data base, the single car comprising specific background image of the different automobile types of preparation
The vehicle target image of image >=100, thus complete the production the vehicle target image data base comprising specific background image,
To meet the needs of deep learning neural network is to image data base；
5th step, with deep learning neural network to the vehicle target image data base comprising specific background image of preparation
It is verified:
To the vehicle target of the single car image of the different automobile types comprising specific background image of above-mentioned 4th step production
Image data base, further transformation includes the size of the vehicle target image of specific background image, as deep learning nerve net
Network input, recognition capability of the verifying deep learning neural network to preparation variable background vehicle target image data base；
So far, the vehicle target image data base comprising specific background image is completed to establish.
The above-mentioned vehicle target image data base method for building up comprising specific background image, the different automobile types are cross-country
Vehicle, car, bus and minibus these fourth types vehicle.
The above-mentioned vehicle target image data base method for building up comprising specific background image, the width of the offroad vehicle are
1.9 meters, the width of car is 1.7 meters, and the width of bus is 2.5 meters, and the width of minibus is 1.5 meters.
The above-mentioned vehicle target image data base method for building up comprising specific background image, the width in the established standards lane
Degree is 3.75 meters.
The above-mentioned vehicle target image data base method for building up comprising specific background image, is equally applicable to offroad vehicle, sedan-chair
Other vehicles except vehicle, bus and minibus these fourth types vehicle.
The above-mentioned vehicle target image data base method for building up comprising specific background image, it is described to pass through image processing software
Vehicle target background image-erasing will be acquired, is the mapping software carried in Windows used herein of image processing software.
The above-mentioned vehicle target image data base method for building up comprising specific background image, the deep learning neural network
It is deep learning neural network LeNet-5.
The above-mentioned vehicle target image data base method for building up comprising specific background image, related operating method are these
Well known to technical field.
The beneficial effects of the present invention are: compared with prior art, substantive distinguishing features outstanding of the invention and marked improvement
It is as follows:
(1) the method for the present invention uses specific background image and vehicle target image synthetic method, acquires specific back respectively
Scape image and vehicle target image, the two is mutually indepedent, therefore selects different specific background images according to demand, acquisition system
Make specific background image and vehicle target image easily synthesize the vehicle target image data base comprising specific background image,
Without repeatedly acquiring the video or image of background in different real backgrounds, and primary collection of material as needed is specific
Background image prepares the vehicle target image data base comprising different background image, therefore the method for the present invention has easy expand
And feature at low cost.
(2) prior art can only determine vehicle by practical scenery in acquisition automobile video frequency and image and then method of cutting out
The background image of target image, cannot change background image, and the method for the present invention is embedded in different under identical specific background image
Vehicle target image, background image indicate that the image of actual area is constant, are equivalent under fixed focal length vehicle of shooting with video-corder or take pictures
Target image, while can arbitrarily change background image, expand vehicle target image under different background image, it is easy to expand vehicle
The capacity in destination image data library.
(3) establish comprising specific background image vehicle target image data base it is existing in the related technology, through frequently with
Existing disclosed vehicle target database, typical such as CompCars and COCO, these databases are third party's acquisitions, existing in shooting
It prepares and is formed under the background image of field, it is impossible to the background image comprising each area and all seasons, in different background figure
The recognition capability that will affect thus manufactured vehicle target database deep learning neural network as under, the present invention overcomes
The defect of another specific region is promoted the use of using certain areas and specific time acquisition vehicle target image, using can change
The specific background image of change, synthesis include the vehicle target image data base of specific background image, are adapted to different regions and not
With the background image in season, there is very strong generalization.
(4) other areas cannot be directly generalized in a regional collection in worksite video and the background image taken pictures,
If acquiring automobile video frequency to all vehicle monitoring points, needs to put into huge fund and huge workload, the present invention and adopt
With specific background image and vehicle target image is independently acquired, image synthetic method is then utilized, simplifies foundation packet
The workload of the vehicle target image data base of the image containing specific background, reduces costs.
(5) it acquires automobile video frequency and then image is cut in the prior art, collect same position vehicle target figure
As needing multitude of video, different vehicle difficult to realize is completely coincident the image in same position, and the method for the present invention is by vehicle mesh
Logo image is embedded in specific background image, it is easy to realize the different vehicle target image of same position.
(6) present invention is using real road width and the pixel to imaged database specific background image bottom, really
Determining specific background image bottom indicates developed width, determines to imaged database background image pass corresponding with practical scenery
System, really reflects the spatial position indicated in specific background image.
(7) present invention indicates developed width, adjustment using practical typical vehicle width combination specific background image bottom
Vehicle target picture traverse, and the height of vehicle target image is adjusted in turn, realize vehicle target image and specific background image
Corresponding relationship.
(8) with individually deep learning neural metwork training is used for using the vehicle target image data base without background image
It compares, the vehicle target image data base comprising specific background image that the present invention establishes more connects due to containing background image
The identification of nearly actual vehicle target.
(9) present invention includes the vehicle target image data base method for building up of specific background image, and preparation includes specific
The vehicle target image data base of background image can be generalized to other network instructions other than for deep learning neural network
White silk, vehicle target image characteristics extraction and identification.
(10) method that CN106846813A constructs urban road vehicle image data base is turned using actual acquisition containing video
It changes image method into, is determined according to these factors of vehicle type, different scenes, shooting angle, illumination variation and weather condition
Data acquisition plan simultaneously acquires road vehicle original video, and original video is switched to image, is delimited in image using ladder-shaped frame
Area-of-interest carries out Fuzzy processing to the regions of non-interest other than ladder-shaped frame, marks the vehicle figure in area-of-interest
Picture obtains vehicle database；The present invention uses specific background and vehicle target image synthetic method, acquires specific background image
The vehicle target image data base comprising specific background image is easily synthesized with vehicle target image, without in specific background
Video or image are acquired again, and background image can change, prepare vehicle destination image data library under different background image,
Have the characteristics that it is easy expansion and it is at low cost.
(11) a kind of vehicle positioning method based on deep neural network image recognition of CN109446973A is not share the same light
According to and weather condition under multiple periods in carry out image pattern acquisition, the present invention use specific background image and vehicle target figure
As synthetic method, acquires specific background image and vehicle target image easily synthesizes the vehicle mesh comprising specific background image
Logo image database, without acquiring video or image again in specific background image, and background image can change, preparation
Vehicle target image data base under different background image, have the characteristics that be easy expand and it is at low cost.
(12) a kind of vehicle front target identification method based on convolutional neural networks of CN107633220A needs to obtain big
Vehicular traffic associated picture is measured as sample, mirror imageization transformation in left and right is carried out to the image of collection, EDS extended data set simultaneously makes mark
Label；The present invention uses specific background image and vehicle target image synthetic method, acquires specific background image and vehicle target
Image easily synthesizes the vehicle target image data base comprising specific background image, without acquiring again in specific background image
Video or image, and background image can change, and prepare vehicle destination image data library under different background image, has and holds
Easily expansion and feature at low cost.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is the operating process schematic block diagram of the method for the present invention.
Fig. 2 is the schematic diagram of specific background image collected in the embodiment of the present invention.
Fig. 3 is the schematic diagram of the offroad vehicle target image collected without background image in the embodiment of the present invention.
Fig. 4 is showing for the offroad vehicle target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
It is intended to.
Fig. 5 is the schematic diagram of the car target image collected without background image in the embodiment of the present invention.
Fig. 6 is the signal of the car target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
Figure.
Fig. 7 is the schematic diagram of the bus target image collected without background image in the embodiment of the present invention.
Fig. 8 is showing for the bus target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
It is intended to.
Fig. 9 is the schematic diagram of the minibus target image collected without background image in the embodiment of the present invention.
Figure 10 is prepared in the embodiment of the present invention to show containing minibus target image under specific background image shown in Fig. 2
It is intended to.
Figure 11 be in the embodiment of the present invention deep learning neural metwork training and test it is prepared include specific background figure
The accuracy rate curve of the vehicle target image data base of picture.
In figure, 201. expressway roadside trees, 202. high speed cement crash bearers, 203. lanes boundary solid line, 204. high speeds
Driving lane (I), 205. lanes boundary dotted line, 206. high speed traveling lanes (II), 207. high-speed isolated bands and trees, 401. get over
The schematic diagram of offroad vehicle target image, the insertion of 601. car target images are high in wild vehicle target image insertion expressway background image
The schematic diagram of car target image in fast road background image, bus target image shows in 801. insertion expressway background images
It is intended to, 1001. minibus target images are embedded in the schematic diagram of minibus target image in the background image of expressway, 1101. training
Collection is the deep learning neural network recognization rate that the every class 140 of four class vehicles opens vehicle target image, and 1102. test sets are four class vehicles
Every class 60 opens the deep learning neural network recognization rate of vehicle target image.
Specific embodiment
Embodiment illustrated in fig. 1 shows that operating process of the invention is: collection of material specific background image calculates specific back
Developed width → production that scape image bottom unit pixel indicates only includes the vehicle target image of single car image and counts
It calculates vehicle width corresponding pixel points → adjustment vehicle target image size vehicle target image is embedded in specific background image
→ the step of repeating above-mentioned second step to third step, vehicle target image data base of the production comprising specific background image → use depth
Degree learning neural network verifies the vehicle target image data base comprising specific background image of preparation.
Embodiment illustrated in fig. 1 shows that the method for the present invention is independent acquisition specific background image and vehicle target image, first adopts
Collection production specific background image, rear production only include the vehicle target image of single car image, then will only include single car
Vehicle target image be embedded into formed in specific background image include specific background image vehicle target image, production includes
The vehicle target image data base of specific background image includes specific background image to preparation with deep learning neural network
Vehicle target image data base is verified, and prepares and meets the needs of deep learning neural network is to image data base and use
The vehicle target image data base comprising specific background image that deep learning neural network is verified.
Fig. 2 is the schematic diagram of specific background image collected in the embodiment of the present invention, and the real sample of the figure is the coloured silk of outdoor scene
Color photo or the effect picture that edge is extracted by greyscale transformation.The trees of roadside containing expressway 201, high speed cement are anti-in the figure image
Hit pier 202, lane boundary solid line 203, high speed traveling lane (I) 204, lane boundary dotted line 205, high speed traveling lane (II)
206, high-speed isolated band and trees 207.This Fig. 2 is the effect picture signal that specific background image extracts edge by greyscale transformation
Figure, prepares four kinds of vehicle vehicle target image data bases under this specific background, and a certain vehicle is prepared with 200 kinds of vehicle targets
Image.In order to draw conveniently, which has been adjusted to 128 × 128 pixels, in the present invention is implemented, the image
Included in constitution element be not to be limited.
Schematic diagram of the Fig. 3 for the offroad vehicle target image collected in the embodiment of the present invention without background image, the figure
Real sample be outdoor scene photochrome or by greyscale transformation extract edge effect picture.In the figure, shown cross-country vehicle-width exists
1.9 meters, any background image is not included in figure in offroad vehicle image.The original background for including for acquisition offroad vehicle target image
Image is deleted its original background image using image processing software.Only be free of the single off-road vehicle of former background image
Target image can be just synthesized in above-mentioned Fig. 2 specific background image, and otherwise former background image can be with above-mentioned Fig. 2 specific background figure
As mixing.This Fig. 3 is the schematic diagram for the effect picture that offroad vehicle vehicle target image extracts edge by greyscale transformation, vehicle
Target image transverse width has been adjusted to 128 pixels.
Fig. 4 is showing for the offroad vehicle target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
It is intended to, the real sample of the figure is the photochrome of outdoor scene or extracts the effect picture at edge by greyscale transformation.The figure is with above-mentioned Fig. 2
Specific background image, the image of specific background containing Fig. 2 being prepared into using offroad vehicle image shown in above-mentioned Fig. 3 as vehicle target image
Offroad vehicle target image, this Fig. 4 are the effect pictures that edge is extracted by greyscale transformation.In figure, offroad vehicle target image is occupied
The right side lane of observer, part is replaced by offroad vehicle image in the specific background image of above-mentioned Fig. 2, the front wheel landing of offroad vehicle
For tire on the road of specific background image bottom, the offroad vehicle vehicle target image for realizing Fig. 3 is embedded in the specific back of Fig. 2
In scape image, the vehicle target image comprising specific background image is made.Offroad vehicle target image shown in Fig. 4 is embedded in expressway
In the schematic diagram 401 for extracting the offroad vehicle target image effect picture at edge in background image by greyscale transformation, road containing expressway
Side trees 201, high speed cement crash bearer 202, lane boundary solid line 203, lane boundary dotted line 205, high speed traveling lane (II)
206, high-speed isolated band and trees 207.
Fig. 5 is the schematic diagram of the car target image collected without background image in the embodiment of the present invention, the figure
Real sample is the photochrome of outdoor scene or extracts the effect picture at edge by greyscale transformation.The present invention is by general car and sport car shape
It is all classified as car model, in the Fig. 5, shown car width is at 1.7 meters.Any background image is not included in Fig. 5 in car image.
Original background image that acquisition car target image includes is deleted its original background image using image processing software.
This Fig. 5 is vehicle mesh in the schematic diagram for the effect picture that the car model target image without background extracts edge by greyscale transformation
The transverse width of logo image has been adjusted to 128 pixels.
Fig. 6 is the signal of the car target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
Figure, the real sample of the figure are the photochrome of outdoor scene or extract the effect picture at edge by greyscale transformation.The figure is spy with above-mentioned Fig. 2
Background image is determined, using the car for the image of specific background containing Fig. 2 that car image shown in above-mentioned Fig. 5 is prepared into as vehicle target image
Target image, this Fig. 6 are the effect pictures that edge is extracted by greyscale transformation.In this Fig. 6, car image is occupied on the right side of observer
Face lane, part is replaced by car image in the specific background image of above-mentioned Fig. 2, and the front wheel landing tire of car is in specific background
On the road of image bottom, the car vehicle target image for realizing Fig. 5 is embedded in the specific background image of Fig. 2, and packet is made
The vehicle target image of the image containing specific background.Compared with offroad vehicle, car height is low, is preparing above-mentioned Fig. 2 specific background figure
As in the lower car vehicle target image comprising Fig. 5, car vehicle target image region in Fig. 2 background is less than offroad vehicle, screening
Region flat in background image is kept off.Become in car target image insertion expressway background image shown in fig. 6 by gray scale
It changes in the car target image effect picture 601 for extracting edge, the trees of roadside containing expressway 201, high speed cement crash bearer 202, vehicle
Road boundary solid line 203, lane boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees 207.
Fig. 7 is the schematic diagram of the bus target image collected without background image in the embodiment of the present invention.The figure
Real sample be outdoor scene photochrome or by greyscale transformation extract edge effect picture.In the figure, public transport vehicle-width is 2.5
Rice.Any background is not included in Fig. 7 in bus image.It include original Background for acquisition public transit vehicle target image
Picture is deleted its original background image using image processing software.This Fig. 7 is the public transit vehicle target image warp without background
The effect picture that edge is extracted in greyscale transformation is crossed, the transverse width of vehicle target image has been adjusted to 128 pixels.
Fig. 8 is showing for the bus target image containing specific background image shown in Fig. 2 prepared in the embodiment of the present invention
It is intended to.The real sample of the figure is the photochrome of outdoor scene or extracts the effect picture at edge by greyscale transformation.The figure is with above-mentioned Fig. 2
Specific background image, the image of specific background containing Fig. 2 being prepared into using bus image shown in above-mentioned Fig. 7 as vehicle target image
Bus target image, this Fig. 8 are the effect pictures that edge is extracted by greyscale transformation.In this Fig. 8, bus image, which occupies, to be seen
The right side lane for the person of examining, part is replaced by bus image in the specific background image of above-mentioned Fig. 2, the front wheel landing wheel of bus
Tire on the road of specific background image bottom, realize with bus image under Fig. 2 specific background image, compare offroad vehicle
And car, bus is bulky, when under preparing above-mentioned Fig. 2 specific background image including bus target image, bus
Target image region in Fig. 2 background is greater than offroad vehicle and car, blocks scenery maximum in former background, occupies most of background
Image.The bus target image effect picture at edge is extracted in insertion expressway background image shown in Fig. 8 by greyscale transformation
In 801, the trees of roadside containing expressway 201, high speed cement crash bearer 202, lane boundary solid line 203, high speed traveling lane (II)
206, high-speed isolated band and trees 207.
Fig. 9 is the schematic diagram of the minibus target image collected without background image in the embodiment of the present invention.The figure
Real sample be outdoor scene photochrome or by greyscale transformation extract edge effect picture.In the figure, bread vehicle-width is 1.5
Rice.It include original background image for acquisition minibus target image, using image processing software, by its original Background
As deleting.This Fig. 9 is the effect picture that the bread vehicle target image without background extracts edge by greyscale transformation, vehicle target
The transverse width of image has been adjusted to 128 pixels.
Figure 10 is prepared in the embodiment of the present invention to show containing minibus target image under specific background image shown in Fig. 2
It is intended to.The real sample of the figure is the photochrome of outdoor scene or extracts the effect picture at edge by greyscale transformation.The figure is with above-mentioned Fig. 2
Specific background image, the image of specific background containing Fig. 2 being prepared into using minibus image shown in above-mentioned Fig. 9 as vehicle target image
Minibus target image, this Figure 10 are the effect pictures that edge is extracted by greyscale transformation.In this Figure 10, minibus image is occupied
The right side lane of observer, part quilt cover hired car image replaces in the specific background image of above-mentioned Fig. 2, the front wheel landing of minibus
Tire on the road of specific background image bottom, realize with minibus image under Fig. 2 specific background image, compared to cross-country
Vehicle, car and bus, minibus is most narrow, includes minibus target image under the 2 specific background images that chart, in Fig. 2 background
Middle elongated zones, the scenery for blocking background are less.It is passed through in minibus target image insertion expressway background image shown in Fig. 10
It crosses greyscale transformation to extract in the minibus target image effect picture 1001 at edge, the trees of roadside containing expressway 201, high speed cement are anti-
Hit pier 202, lane boundary solid line 203, lane boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees
207。
Figure 11 be in the embodiment of the present invention deep learning neural metwork training and test it is prepared include specific background figure
The accuracy rate curve of the vehicle target image data base of picture.Embodiment is as it can be seen that the present invention forms 200 images as shown in Figure 11
Thus every one kind prepares composition and includes offroad vehicle, car, bus and the minibus four under specific background image shown in Fig. 2
The vehicle target image data base of class vehicle.The vehicle target image data base comprising specific background image prepared,
After size is converted to 32 × 32 pixels, it is used for exemplary depth learning neural network LeNet-5, wherein a certain type of vehicle has
200 images, selection wherein are used to train depth network for 140, and in addition 60 are used to test, deep learning neural network
It is 2 that LeNet-5, which learns rate coefficient, and training uses 20 image samples every time, the knowledge of training set and test set after training 150 times
Other accuracy has been above 95%, can be realized the training to deep learning neural network structure.In Figure 11, curve 1101 is instruction
Practicing collection is the deep learning neural network recognization rate that the every class 140 of four class vehicles opens vehicle target image, and curve 1102 is test set
The deep learning neural network recognization rate of vehicle target image is opened for the every class 60 of four class vehicles.
Embodiment 1
The vehicle of the vehicle of the vehicle target image of institute's collection of material includes offroad vehicle, car, bus in the present embodiment
And minibus.
The vehicle target image data base method for building up comprising specific background image of the present embodiment, the specific steps are as follows:
The first step, collection of material specific background image:
Pass through the method or network download method taken pictures, collection of material specific background image, institute's collection of material specific background
Image request is as follows:
(1.1) picture material included in specific background image:
Acquire picture material included in specific background image are as follows:
Picture material included in specific background image is that carriageway image adds Tree image, building object image, mountain range figure
1~4 kind in picture and road sign image；
It include expressway road in picture material included in specific background image as shown in Fig. 2 in the present embodiment
Side trees 201, high speed cement crash bearer 202, lane boundary solid line 203, high speed traveling lane (I) 204, lane boundary dotted line
205, high speed traveling lane (II) 206, high-speed isolated band and trees 207；
Offroad vehicle: pass through the specific background image for the method collection of material taken pictures, as shown in Fig. 4, specific background image
Included in picture material, comprising expressway roadside trees 201, high speed cement crash bearer 202, lane boundary solid line 203,
Lane boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees 207；
Car: by the specific background image for the method collection of material taken pictures, as shown in Fig. 6, in specific background image
In the picture material for being included, include expressway roadside trees 201, high speed cement crash bearer 202, lane boundary solid line 203, vehicle
Road boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees 207；
Bus: pass through the specific background image for the method collection of material taken pictures, as shown in Fig. 8, specific background image
Included in picture material, comprising expressway roadside trees 201, high speed cement crash bearer 202, lane boundary solid line 203,
High speed traveling lane (II) 206, high-speed isolated band and trees 207；
Minibus: pass through the specific background image for the method collection of material taken pictures, as shown in Fig. 10, specific background image
Included in picture material, comprising expressway roadside trees 201, high speed cement crash bearer 202, lane boundary solid line 203,
Lane boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees 207；
(1.2) the pixel size of specific background image:
Specific background image slices vegetarian refreshments size is N pixel × M pixel, i.e., it is laterally M pixel, specific back that longitudinal, which is N pixel,
The vehicle target image slices vegetarian refreshments to be prepared for scape image slices vegetarian refreshments >=included by background image；
Specific background image slices vegetarian refreshments size in the present embodiment is N × M=128 pixel × 128 pixels, i.e., longitudinal is N
=128 pixels are laterally M=128 pixel, the vehicle to be prepared for specific background image slices vegetarian refreshments >=included by background image
Target image pixel；
(1.3) developed width that specific background image bottom unit pixel indicates is calculated:
The width in established standards lane is L meters, according to the occupied I pixel in lane in specific background image bottom, with such as
Lower formula (1) calculates the developed width R that the bottom unit pixel of specific background image indicates,
In above-mentioned formula (1), the unit of R is rice/pixel；
In the present embodiment,
Offroad vehicle, car, bus and minibus are identical: the width in established standards lane is L=3.75 meters, according to specific
The occupied I=128 pixel in lane in background image bottom substitutes into L=3.75 meters and I=128 pixel, obtains R=
0.029297 (rice/pixel)；
Second step, production only include the vehicle target image of single car image:
By taking pictures or network download method, acquire and make only include single car image a vehicle target image, should
Only the requirement of the vehicle target image comprising single car image is as follows:
(2.1) direct picture of simple target vehicle is acquired；
(2.2) according to the size and shape of the vehicle of institute's collection of material vehicle target image, it is divided into different automobile types, this implementation
It include offroad vehicle, car, bus and minibus in example；
(2.3) when take pictures or the vehicle target image of network downloading in comprising corresponding original background image when, pass through figure
It so only include simple target in the vehicle target image for acquiring and converting as processing software deletes the original background image
Vehicle image, the vehicle width in the vehicle target image are J pixel, acquire the vehicle width pixel J in vehicle target image
It is greater than the occupied I pixel in lane in 0.9 times of specific background image bottom；
In the present embodiment,
Offroad vehicle: the vehicle target image comprising single offroad vehicle vehicle image is as shown in Fig. 3, and the real sample of the figure is real
The photochrome of scape passes through the effect picture that edge is extracted in greyscale transformation, and the vehicle width in the offroad vehicle vehicle target image is
J=128 pixel meets the requirement of the occupied I=128 pixel in lane in specific background image bottom of the J greater than 0.9 times；
Car: the vehicle target image comprising single car vehicle image is as shown in Fig. 5, and the real sample of the figure is outdoor scene
Photochrome passes through the effect picture that edge is extracted in greyscale transformation, and the vehicle width in the car vehicle target image is J=128
Pixel meets the requirement of the occupied I=128 pixel in lane in specific background image bottom of the J greater than 0.9 times；
Bus: the vehicle target image comprising single bus vehicle image is as shown in Fig. 7, and the real sample of the figure is real
The photochrome of scape passes through the effect picture that edge is extracted in greyscale transformation, and the vehicle width in the bus vehicle target image is
J=128 pixel meets the requirement of the occupied I=128 pixel in lane in specific background image bottom of the J greater than 0.9 times；
Minibus: the vehicle target image comprising single minibus vehicle image is as shown in Fig. 9, and the real sample of the figure is real
The photochrome of scape passes through the effect picture that edge is extracted in greyscale transformation, and the vehicle width in the minibus vehicle target image is
J=128 pixel meets the requirement of the occupied I=128 pixel in lane in specific background image bottom of the J greater than 0.9 times；
(2.4) the corresponding pixel of vehicle width is calculated:
According to different automobile types described in above-mentioned (2.2) step, measuring its width is W meters, is calculated with following formula (2) corresponding
The vehicle laterally corresponds to the pixel D of the specific background image of above-mentioned first step collection of material,
In formula (2), fix () indicates to be rounded；
In the present embodiment,
Offroad vehicle: cross-country vehicle-width W=1.9 meters and R=0.029297 (rice/pixel) are substituted into, obtain D=65 pixel, more
D=65 pixel is occupied in bottom in wild 2 specific background image of vehicle transverse direction corresponding diagram；
Car: substituting into car width W=1.7 meters and R=0.029297 (rice/pixel), obtains D=58 pixel, and car is horizontal
Into 2 specific background image of corresponding diagram, D=58 pixel is occupied in bottom；
Bus: substituting into W=2.5 meters of public transport vehicle-width and R=0.029297 (rice/pixel), obtain D=85 pixel, public
Bottom in 2 specific background image of vehicle transverse direction corresponding diagram is handed over to occupy D=85 pixel；
Minibus: W=1.5 meters of bread vehicle-width and R=0.029297 (rice/pixel) are substituted into, D=51 pixel, face are obtained
It wraps bottom in 2 specific background image of lateral corresponding diagram and occupies D=51 pixel；
Vehicle target image is embedded in specific background image by third step:
The only vehicle target image insertion above-mentioned first comprising single offroad vehicle vehicle image that above-mentioned second step is made
In the specific background image for walking collection of material, the specific operation method is as follows:
(3.1) size of vehicle target image is adjusted:
The vehicle mesh comprising single offroad vehicle vehicle image of above-mentioned second step production is calculated separately with following formula (3)
The horizontal and vertical regulation coefficient E of logo image, i.e., horizontal and vertical adjust separately of single offroad vehicle vehicle target image is E times
The size that vehicle target image is adjusted according to the numerical value of regulation coefficient E, the vehicle target image after being sized
For in the vehicle target image insertion specific background image of next step；
In the present embodiment,
Offroad vehicle: D=65 pixel, J=128 pixel, E=0.5078, E < 1, reduce the production of above-mentioned second step only includes
The vehicle target image of single offroad vehicle vehicle image is made correspondingly sized to be adjusted to original image respectively by attached drawing 2 is horizontal and vertical
0.5078 times of picture reduces the size of the offroad vehicle target image of acquisition, and the vehicle target image after being sized is used for down
In the vehicle target image insertion specific background image of one step；
Car: D=58 pixel, J=128 pixel, E=0.4531, E < 1, reduce above-mentioned second step production only includes single
The vehicle target image of one offroad vehicle vehicle image is made correspondingly sized to be adjusted to original image respectively by attached drawing 2 is horizontal and vertical
0.4531 times, reduce the size of the offroad vehicle target image of acquisition, the vehicle target image after being sized is used for next
In the vehicle target image insertion specific background image of step；
Bus: D=85 pixel, J=128 pixel, E=0.6641, E < 1, reduce the production of above-mentioned second step only includes
The vehicle target image of single offroad vehicle vehicle image is made correspondingly sized to be adjusted to original image respectively by attached drawing 2 is horizontal and vertical
0.6641 times of picture reduces the size of the offroad vehicle target image of acquisition, and the vehicle target image after being sized is used for down
In the vehicle target image insertion specific background image of one step；
Minibus: D=51 pixel, J=128 pixel, E=0.3984, E < 1, reduce the production of above-mentioned second step only includes
The vehicle target image of single offroad vehicle vehicle image is made correspondingly sized to be adjusted to original image respectively by attached drawing 2 is horizontal and vertical
0.3984 times of picture reduces the size of the offroad vehicle target image of acquisition, and the vehicle target image after being sized is used for down
In the vehicle target image insertion specific background image of one step；
It (3.2) will be in the specific background image of the vehicle target image insertion collection of material only comprising single car image:
The carriageway image near center location shown in the specific background image of above-mentioned first step collection of material, in insertion
(3.1) step adjusted size of vehicle target image is stated, which is placed on above-mentioned first step collection of material
Position more than specific background image bottom, the vehicle target image front wheel landing, tire is occupied to be acquired in the above-mentioned first step
Lane position in the specific background image of production, which is opaque；
(3.3) size according to lane in the attached specific background image shown in Fig. 2 of above-mentioned first step collection of material determines
The developed width that the specific background picture traverse pixel indicates, combining target vehicle width determine that the vehicle target image occupies spy
Determine the quantity of horizontal pixel in background image；
(3.4) adjustment only includes the quantity of the vehicle target image slices vegetarian refreshments of single offroad vehicle vehicle image, to keep this
The laterally and longitudinally same ratio of vehicle target image changes, and only changes the size of the vehicle target image, and vehicle shape is kept
It is constant；
Thus the vehicle target image of the single car image comprising specific background image is made；
In the present embodiment,
Offroad vehicle: it is as shown in Fig. 4 to be thus made the offroad vehicle vehicle target image comprising specific background image, the figure with
Above-mentioned Fig. 2 is specific background image, is prepared into using offroad vehicle image shown in above-mentioned Fig. 3 as vehicle target image specific containing Fig. 2
The offroad vehicle target image of background image, this Fig. 4 are the effect pictures that edge is extracted by greyscale transformation.In figure, offroad vehicle target
Image occupies the part in the right side lane of observer, the specific background image of above-mentioned Fig. 2 and is replaced by offroad vehicle image, offroad vehicle
Front wheel landing tire on the road of specific background image bottom, realize Fig. 3 offroad vehicle vehicle target image insertion
In the specific background image of Fig. 2, the vehicle target image comprising specific background image is made.Offroad vehicle target figure shown in Fig. 4
In offroad vehicle target image effect picture 401 as extracting edge in insertion expressway background image by greyscale transformation, containing high speed
Road roadside trees 201, high speed cement crash bearer 202, lane boundary solid line 203, lane boundary dotted line 205, high speed traveling lane
(II) 206, high-speed isolated band and trees 207；
Car: thus it is made that the car vehicle target image comprising specific background image is as shown in Fig. 6, and the figure is with above-mentioned
Fig. 2 is specific background image, the figure of specific background containing Fig. 2 being prepared into using car image shown in above-mentioned Fig. 5 as vehicle target image
The car target image of picture, this Fig. 6 are the effect pictures that edge is extracted by greyscale transformation.In this Fig. 6, car image, which occupies, to be seen
The right side lane for the person of examining, part is replaced by car image in the specific background image of above-mentioned Fig. 2, and the front wheel landing tire of car exists
On the road of specific background image bottom, the car vehicle target image for realizing Fig. 5 is embedded in the specific background image of Fig. 2
In, the vehicle target image comprising specific background image is made.Compared with offroad vehicle, car height is low, special preparing above-mentioned Fig. 2
Determine in the car vehicle target image under background image comprising Fig. 5, car vehicle target image region in Fig. 2 background is less than more
Wild vehicle has blocked region flat in background image.Pass through in car target image insertion expressway background image shown in fig. 6
Greyscale transformation is extracted in the car target image effect picture 601 at edge, the trees of roadside containing expressway 201, high speed cement crash bearer
202, lane boundary solid line 203, lane boundary dotted line 205, high speed traveling lane (II) 206, high-speed isolated band and trees 207；
Bus: it is as shown in Fig. 8 to be thus made the bus vehicle target image comprising specific background image, the figure with
Above-mentioned Fig. 2 is specific background image, is prepared into using bus image shown in above-mentioned Fig. 7 as vehicle target image specific containing Fig. 2
The bus target image of background image, this Fig. 8 are the effect pictures that edge is extracted by greyscale transformation.In this Fig. 8, bus figure
Replaced as occupying the part in the right side lane of observer, the specific background image of above-mentioned Fig. 2 by bus image, bus
Front wheel landing tire on the road of specific background image bottom, realize with bus image under Fig. 2 specific background image,
Compared to offroad vehicle and car, bus is bulky, includes bus target image in the case where preparing above-mentioned Fig. 2 specific background image
When, bus target image region in Fig. 2 background is greater than offroad vehicle and car, blocks scenery maximum in former background, occupies
Most of background image.The bus target at edge is extracted in insertion expressway background image shown in Fig. 8 by greyscale transformation
In image effect Figure 80 1, the trees of roadside containing expressway 201, high speed cement crash bearer 202, lane boundary solid line 203, high speed row
Vehicle lane (II) 206, high-speed isolated band and trees 207；
Minibus；Thus it is as shown in Fig. 10 that the minibus vehicle target image comprising specific background image is made, the figure
Using above-mentioned Fig. 2 as specific background image, it is prepared into using minibus image shown in above-mentioned Fig. 9 as vehicle target image special containing Fig. 2
Determine the minibus target image of background image, this Figure 10 is the effect picture that edge is extracted by greyscale transformation.In this Figure 10, bread
Vehicle image occupies the part quilt cover hired car image in the right side lane of observer, the specific background image of above-mentioned Fig. 2 and replaces, bread
The front wheel landing tire of vehicle realizes and minibus under Fig. 2 specific background image on the road of specific background image bottom
Image compares offroad vehicle, car and bus, and minibus is most narrow, includes minibus target figure under the 2 specific background images that chart
Picture, the elongated zones in Fig. 2 background, the scenery for blocking background are less.Minibus target image shown in Fig. 10 is embedded in expressway
In the minibus target image effect picture 1001 for extracting edge in background image by greyscale transformation, the trees of roadside containing expressway
201, high speed cement crash bearer 202, lane boundary solid line 203, lane boundary dotted line 205, high speed traveling lane (II) 206, height
Fast isolation strip and trees 207；
4th step, production include the vehicle target image data base of specific background image:
The step of repeating above-mentioned second step to third step prepares comprising specific background image all different automobile types
Single car image vehicle target image data base, the single car comprising specific background image of the different automobile types of preparation
The vehicle target image of image >=100, thus complete the production the vehicle target image data base comprising specific background image,
To meet the needs of deep learning neural network is to image data base；
Different automobile types in the present embodiment are offroad vehicle, car, bus, minibus；
5th step, with deep learning neural network to the vehicle target image data base comprising specific background image of preparation
It is verified:
To above-mentioned 4th step production the offroad vehicle comprising specific background image, car, bus, minibus single vehicle
The vehicle target image data base of image, further the size of vehicle target image of the transformation comprising specific background image, makees
For the input of deep learning neural network, verifying deep learning neural network is to preparation variable background vehicle target image data base
Recognition capability；
In the present embodiment, as shown in Figure 11,200 images are formed every one kind by the present embodiment, and thus preparing composition includes
The vehicle target image data of offroad vehicle, car, bus and four class vehicle of minibus under specific background image shown in Fig. 2
Library.The vehicle target image data base comprising specific background image prepared, after size is converted to 32 × 32 pixels,
For exemplary depth learning neural network LeNet-5, wherein a certain type of vehicle has 200 images, selection is wherein used for for 140
Training depth network, in addition 60 are used to test, and it is 2 that deep learning neural network LeNet-5, which learns rate coefficient, and training is adopted every time
With 20 image samples, the recognition correct rate of training set and test set has been above 95% after training 150 times, can be realized pair
The training of deep learning neural network structure.In Figure 11, curve 1101 is that training set is that the every class 140 of four class vehicles opens vehicle target
The deep learning neural network recognization rate of image, curve 1102 are that test set is that the every class 60 of four class vehicles opens vehicle target image
Deep learning neural network recognization rate.
So far, the vehicle target image data base comprising specific background image is completed to establish.
1. lane of table and four type of vehicle representative widths and corresponding prepare occupy pixel and regulation coefficient in image
Embodiment 2
The picture material included in the specific background image is other same embodiments in addition to carriageway image adds road sign image
1.Embodiment 3
The picture material included in the specific background image is in addition to carriageway image adds road sign image and building figure,
He is the same as embodiment 1.
Embodiment 4
Except picture material included in specific background image is that carriageway image adds Tree image, building object image, mountain range
Except image and road sign image, other are the same as embodiment 1.
It is described that vehicle target background image-erasing will be acquired by image processing software in above-described embodiment, it is used here
Image processing software be in Windows carry mapping software；The deep learning neural network is deep learning nerve net
Network LeNet-5；Related operating method is well known to the art.
Claims (4)
1. including the vehicle target image data base method for building up of specific background image, it is characterised in that: prepared includes spy
The vehicle target image data base for determining background image will use deep learning neural network to be verified, the specific steps are as follows:
The first step, collection of material specific background image:
By taking pictures or network download method, collection of material specific background image, institute's collection of material specific background image request is such as
Under:
(1.1) picture material included in specific background image:
Picture material included in specific background image is, carriageway image add Tree image, building object image, mountain range image and
1~4 kind in road sign image；
(1.2) the pixel size of specific background image:
Specific background image slices vegetarian refreshments size is N pixel × M pixel, i.e., it is laterally M pixel, specific background figure that longitudinal, which is N pixel,
As the vehicle target image slices vegetarian refreshments to be prepared for pixel >=included by background image；
(1.3) developed width that specific background image bottom unit pixel indicates is calculated:
The width in established standards lane is L meters, according to the occupied I pixel in lane in specific background image bottom, with following public affairs
Formula (1) calculates the developed width R that the bottom of specific background image is indicated with unit pixel,
In above-mentioned formula (1), the unit of R is rice/pixel；
Second step, production only include the vehicle target image of single car image:
By taking pictures or network download method, acquire and make only include single car image a vehicle target image, this packet
The requirement of the vehicle target image of the image containing single car is as follows:
(2.1) direct picture of simple target vehicle is acquired；
(2.2) according to the size and shape of the vehicle of institute's collection of material vehicle target image, it is divided into different automobile types；
(2.3) when take pictures or the vehicle target image of network downloading in comprising corresponding original background image when, at image
Reason software deletes the original background image, so only includes simple target vehicle in the vehicle target image for acquiring and converting
Image, the vehicle width in the vehicle target image are J pixel, and the vehicle width pixel J acquired in vehicle target image is big
The occupied I pixel in lane in 0.9 times of specific background image bottom；
(2.4) vehicle width corresponding pixel points are calculated:
According to different automobile types described in above-mentioned (2.2) step, measuring its width is W meters, and being calculated with following formula (2) mutually should vehicle
Type laterally corresponds to the pixel D of the specific background image of above-mentioned first step collection of material,
In formula (2), fix () indicates to be rounded；
Third step, will be in the vehicle target image insertion specific background image only comprising single car image:
The only vehicle target image comprising single car image that above-mentioned second step is made is embedded in above-mentioned first step collection of material
Specific background image in, the specific operation method is as follows:
(3.1) size of vehicle target image is adjusted:
The only vehicle target image comprising single car image for calculating separately above-mentioned second step production with following formula (3) is lateral
With longitudinal regulation coefficient E,
The size of vehicle target image is adjusted according to the numerical value of regulation coefficient E, the vehicle target image after being sized is used for
In the vehicle target image insertion specific background image of next step；
It (3.2) will be in the specific background image of the vehicle target image insertion collection of material only comprising single car image:
The carriageway image near center location shown in the specific background image of above-mentioned first step collection of material is embedded in above-mentioned
(3.1) adjusted size of vehicle target image, the vehicle target image are placed on the spy of above-mentioned first step collection of material to step
Determine the position of background image bottom or more, the vehicle target image front wheel landing, tire is occupied to acquire in the above-mentioned first step and be made
Lane position in the specific background image of work, which is opaque；
(3.3) size according to lane in the specific background image of above-mentioned first step collection of material determines that the specific background image is wide
The developed width that pixel indicates is spent, combining target vehicle width determines that the vehicle target image occupies in specific background image laterally
The quantity of pixel；
(3.4) adjustment only includes the quantity of the vehicle target image slices vegetarian refreshments of single car image, to keep the vehicle target figure
The laterally and longitudinally same ratio of picture changes, and only changes the size of the vehicle target image, vehicle shape remains unchanged；
Thus the vehicle target image of the single car image comprising specific background image is made；
4th step, production include the vehicle target image data base of specific background image:
The step of repeating above-mentioned second step to third step prepares the list comprising specific background image for all different automobile types
The vehicle target image data base of one vehicle image, the single car image comprising specific background image of the different automobile types of preparation
Vehicle target image >=100, complete the production the vehicle target image data base comprising specific background image, thus with full
Demand of the sufficient deep learning neural network to image data base；
5th step is carried out with the vehicle target image data base comprising specific background image of the deep learning neural network to preparation
Verifying:
To the vehicle target image of the single car image of the different automobile types comprising specific background image of above-mentioned 4th step production
Database, further the size of vehicle target image of the transformation comprising specific background image, defeated as deep learning neural network
Enter, recognition capability of the verifying deep learning neural network to preparation variable background vehicle target image data base；
So far, the vehicle target image data base comprising specific background image is completed to establish.
2. according to claim 1 including the vehicle target image data base method for building up of specific background image, feature exists
In: the different automobile types are offroad vehicle, car, bus and minibus these fourth types vehicle.
3. it according to claim 2 include the vehicle target image data base method for building up of specific background image, it is described cross-country
The width of vehicle is 1.9 meters, and the width of car is 1.7 meters, and the width of bus is 2.5 meters, and the width of minibus is 1.5 meters.
4. according to claim 1 including the vehicle target image data base method for building up of specific background image, feature exists
In: the width in the established standards lane is 3.75 meters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910278860.4A CN110008360B (en) | 2019-04-09 | 2019-04-09 | Method for establishing vehicle target image database containing specific background image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910278860.4A CN110008360B (en) | 2019-04-09 | 2019-04-09 | Method for establishing vehicle target image database containing specific background image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110008360A true CN110008360A (en) | 2019-07-12 |
| CN110008360B CN110008360B (en) | 2020-11-03 |
Family
ID=67170402
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910278860.4A Expired - Fee Related CN110008360B (en) | 2019-04-09 | 2019-04-09 | Method for establishing vehicle target image database containing specific background image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008360B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110443245A (en) * | 2019-08-14 | 2019-11-12 | 上海世茂物联网科技有限公司 | Localization method, device and the equipment of a kind of license plate area under unrestricted scene |
| CN111083362A (en) * | 2019-12-13 | 2020-04-28 | 山东浪潮通软信息科技有限公司 | Method for realizing automatic focusing of vehicle entering and exiting warehouse |
| CN111460186A (en) * | 2020-03-31 | 2020-07-28 | 河北工业大学 | Method for establishing database containing vehicle visible light images and infrared images |
| CN114173059A (en) * | 2021-12-09 | 2022-03-11 | 广州阿凡提电子科技有限公司 | Video editing system, method and device |
| CN117253231A (en) * | 2023-11-15 | 2023-12-19 | 四川弘和数智集团有限公司 | Oil-gas station image processing method and device, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976345A (en) * | 2016-04-27 | 2016-09-28 | 北京航空航天大学 | Visible light remote sensing image synthesis method |
| CN107480611A (en) * | 2017-07-31 | 2017-12-15 | 浙江大学 | A kind of crack identification method based on deep learning convolutional neural networks |
| CN108460414A (en) * | 2018-02-27 | 2018-08-28 | 北京三快在线科技有限公司 | Generation method, device and the electronic equipment of training sample image |
| CN108986058A (en) * | 2018-06-22 | 2018-12-11 | 华东师范大学 | The image interfusion method of lightness Consistency Learning |
| AU2018102037A4 (en) * | 2018-12-09 | 2019-01-17 | Ge, Jiahao Mr | A method of recognition of vehicle type based on deep learning |
| CN109410211A (en) * | 2017-08-18 | 2019-03-01 | 北京猎户星空科技有限公司 | The dividing method and device of target object in a kind of image |
-
2019
- 2019-04-09 CN CN201910278860.4A patent/CN110008360B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105976345A (en) * | 2016-04-27 | 2016-09-28 | 北京航空航天大学 | Visible light remote sensing image synthesis method |
| CN107480611A (en) * | 2017-07-31 | 2017-12-15 | 浙江大学 | A kind of crack identification method based on deep learning convolutional neural networks |
| CN109410211A (en) * | 2017-08-18 | 2019-03-01 | 北京猎户星空科技有限公司 | The dividing method and device of target object in a kind of image |
| CN108460414A (en) * | 2018-02-27 | 2018-08-28 | 北京三快在线科技有限公司 | Generation method, device and the electronic equipment of training sample image |
| CN108986058A (en) * | 2018-06-22 | 2018-12-11 | 华东师范大学 | The image interfusion method of lightness Consistency Learning |
| AU2018102037A4 (en) * | 2018-12-09 | 2019-01-17 | Ge, Jiahao Mr | A method of recognition of vehicle type based on deep learning |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110443245A (en) * | 2019-08-14 | 2019-11-12 | 上海世茂物联网科技有限公司 | Localization method, device and the equipment of a kind of license plate area under unrestricted scene |
| CN110443245B (en) * | 2019-08-14 | 2022-02-15 | 上海世茂物联网科技有限公司 | License plate region positioning method, device and equipment in non-limited scene |
| CN111083362A (en) * | 2019-12-13 | 2020-04-28 | 山东浪潮通软信息科技有限公司 | Method for realizing automatic focusing of vehicle entering and exiting warehouse |
| CN111460186A (en) * | 2020-03-31 | 2020-07-28 | 河北工业大学 | Method for establishing database containing vehicle visible light images and infrared images |
| CN111460186B (en) * | 2020-03-31 | 2022-04-08 | 河北工业大学 | Method for establishing database containing vehicle visible light images and infrared images |
| CN114173059A (en) * | 2021-12-09 | 2022-03-11 | 广州阿凡提电子科技有限公司 | Video editing system, method and device |
| CN117253231A (en) * | 2023-11-15 | 2023-12-19 | 四川弘和数智集团有限公司 | Oil-gas station image processing method and device, electronic equipment and storage medium |
| CN117253231B (en) * | 2023-11-15 | 2024-01-26 | 四川弘和数智集团有限公司 | Oil-gas station image processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110008360B (en) | 2020-11-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110008360A (en) | Vehicle target image data base method for building up comprising specific background image | |
| Chen et al. | Hierarchical object oriented classification using very high resolution imagery and LIDAR data over urban areas | |
| CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
| CN106326858A (en) | Road traffic sign automatic identification and management system based on deep learning | |
| CN103578119B (en) | Target detection method in Codebook dynamic scene based on superpixels | |
| CN110263717B (en) | Method for determining land utilization category of street view image | |
| CN110490802A (en) | A kind of satellite image Aircraft Targets type identifier method based on super-resolution | |
| CN110175576A (en) | A kind of driving vehicle visible detection method of combination laser point cloud data | |
| CN107871126A (en) | Model recognizing method and system based on deep-neural-network | |
| CN108549864A (en) | Area-of-interest filter method towards vehicle-mounted thermal imaging pedestrian detection and device | |
| CN105844257A (en) | Early warning system based on machine vision driving-in-fog road denoter missing and early warning method | |
| US7822224B2 (en) | Terrain map summary elements | |
| CN113255589B (en) | Target detection method and system based on multi-convolution fusion network | |
| CN107122701A (en) | A kind of traffic route sign based on saliency and deep learning | |
| CN107301378A (en) | The pedestrian detection method and system of Multi-classifers integrated in image | |
| CN107045722B (en) | Merge the video signal process method of static information and multidate information | |
| CN105930803A (en) | Preceding vehicle detection method based on Edge Boxes and preceding vehicle detection device thereof | |
| CN104463168B (en) | A kind of useless house site information automation extraction method of sky based on remote sensing image | |
| CN111598048A (en) | Urban village-in-village identification method integrating high-resolution remote sensing image and street view image | |
| CN108052933A (en) | Road Identification system and method based on convolutional neural networks | |
| CN106919939B (en) | A kind of traffic signboard tracks and identifies method and system | |
| CN109697397A (en) | A kind of object detection method and device | |
| JP2012168098A (en) | Photographic position specification system, photographic position specification program, and method for specifying photographic position | |
| CN106951829A (en) | A kind of notable method for checking object of video based on minimum spanning tree | |
| CN106570883A (en) | People counting method based on RGB-D camera |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201103 |