CN108648210A - It is a kind of static state complex scene under fast multi-target detection method and device - Google Patents
It is a kind of static state complex scene under fast multi-target detection method and device Download PDFInfo
- Publication number
- CN108648210A CN108648210A CN201810437311.2A CN201810437311A CN108648210A CN 108648210 A CN108648210 A CN 108648210A CN 201810437311 A CN201810437311 A CN 201810437311A CN 108648210 A CN108648210 A CN 108648210A
- Authority
- CN
- China
- Prior art keywords
- pixel
- background
- code book
- target
- symbol
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30236—Traffic on road, railway or crossing
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
Fast multi-target detection method and device under a kind of static complex scene of the present invention, are related to information monitoring technical field more particularly to a kind of multi-target detection method in photoelectric detecting system, and the device measured using this method.This method is as follows：The image in input video is acquired, background model is established；Each pixel in current frame image is matched with background model, carries out the label of background or foreground；Foreground comprising candidate target, background binary image are handled, false-alarm is removed；The present invention simplifies optimization by being carried out to classical code book model in editable device, in combination in DSP module to object filtering, sequence, the processing of number, solves the disadvantage that conventional method false-alarm is high, verification and measurement ratio is low, flase drop is high, the fast and effective detection to multiple target under static complex scene is realized, preferable effect is achieved in actual scene application.
Description
Technical field
The present invention relates to information monitoring technical field more particularly to a kind of multi-target detections in photoelectric detecting system
Method and device.
Background technology
For still camera, existing technology can accurately detect the foreground moving object under simple background.And
For complicated dynamic background, a variety of interference of dynamic background are contained in scene, leaf, fountain and the light such as shaken
According to variation so that either large or small dynamic change constantly occurs for background in video, greatly reduces the Detection accuracy of the prior art,
A large amount of false-alarms are brought simultaneously, can not accurately be quickly detected target.
In the complex background for shooting dynamic change for stationary cameras, the method for obtaining foreground moving object has very much, packet
Include inter-frame difference, background difference and optical flow method, but that there are accuracy of detection is low for these methods, detection target is imperfect, or even to slow
There is " cavity " phenomenon in the target moved slowly, more even exist and calculate complexity, hardware requirement is high, leads to existing hardware platform
The case where cannot be satisfied, and easily influenced by factors such as noise, illumination variations, it cannot be satisfied actual application.
Invention content
In view of the deficiencies of the prior art, fast multi-target detection method under a kind of static complex scene of present invention offer, with
And the device being detected using this method.
Fast multi-target detection method under a kind of static complex scene of present invention proposition, this method are as follows：
The image in input video is acquired, background model is established；
Each pixel in current frame image is matched with background model, if successful match, which is labeled as background, it is no
Then it is labeled as foreground；
Foreground comprising candidate target, background binary image are handled, false-alarm is removed, obtains the number of candidate target region
And label；
Candidate target region is screened and is sorted, is exported after obtaining the number and number in effective target region；
Fast multi-target detection is completed under static complex scene.
Further, specific implementation is as follows：
At least 1 frame image for acquiring and preserving inputting video data, establishes background model using background modeling algorithm and stores；
Temporarily, each pixel in image to be matched with established background model, if successful match, by this in present frame
Pixel is labeled as background, is otherwise labeled as foreground；
The foreground and background bianry image for including candidate target is obtained, to carrying out cluster point comprising foreground and background bianry image
Analysis, obtains the number and label of candidate target region；
Above-mentioned candidate target region mark is screened and sorted, false-alarm is removed, obtains the number and number in real goal region
After export
Fast multi-target detection is completed under static complex scene.
Further, the method for establishing background model is as follows：
The image of modeling is obtained from inputting video data；
Several frame images are taken to be modeled, wherein it is preferred that 50-200 frame images；
Code book modeling is carried out to each pixel in each frame image, obtains code book model；
Code book model is simplified, all symbols in check image in each pixel code book；If some symbol in code book
Longest not renewal time λ ＞ 50, then be set as 0 by the corresponding flag bit of the symbol；
Complete background modeling.
Further, the method for the code book modeling is as follows：
Initialization code book model is set as sky, gray processing processing is carried out to input picture, is built for each pixel of input picture
A code book is found, the space of each code book of the present invention is 12 symbols, effective marker position is initialized as 0, the study of symbol
Range determines that each code element structure is C according to the reserved number of symbol spacei={ Ymax, Ymin, Ylow, Yhigh, tlast, λ },
Wherein Ymax, YminThe respectively maximum and minimum value of current pixel gray value, Ylow, YhighFor symbol study lower limit and
The upper limit, initial value are the gray value of current pixel；tlastIt is renewal time, initial value 0 for symbol last time matching；λ
For symbol longest not renewal time, initial value 0, the value adds 1 when not updating；
It will be divided into four pieces per frame image, carry out concurrent operation, when a new frame image arrives, the code book time adds 1, by pixel
Gray value is limited to the value range of current grayvalue ［ ± 15, ± 25 ］, and the present invention is preferably ± 20, therefore under the study of code word
It is limited to Ylow- 20, upper limit Yhigh+ 20, wherein code book timing definition is current modeling frame number, and initial value 0, per frame, modeling adds
1；
If the gray value Y of pixel learns in symbol between the upper limit and lower limit, i.e. Ylow≤Y≤Yhigh, then successful match, when code book
Between plus 1, as the following formula the maximin of more new symbol, update the study bound of the symbol：
Ct={ max (Ymax, Y), min (Ymin, Y), Y-20, Y+20, t, λ }；
If the gray value of pixelLearn between the upper limit and lower limit beyond symbol, then it is assumed that pixel does not find matched code word, then
New symbol C is created as the following formulaL, code book series adds 1：
CL={ Ymax, Ymin, Ylow, Yhigh, 0,0 }；
Then the study bound of the symbol is updated, and its effective marker position is set as 1；
By other symbol longests of the code book, renewal time λ does not add 1；
Check that all flag bits are 1 effective code element, if the longest of some symbol not renewal time λ ＞ 50, effective marker position
It is set as 0, modeling terminates.
Further, the code book model of corresponding position pixel carries out in each pixel Yu background model of the present frame
Matching, if the pixel is between code book upper and lower limit in present frame, is denoted as background, is otherwise denoted as foreground；
Matching formula：
Ylow≤Y≤Yhigh；
Wherein Ylow, YhighTo obtain the code book upper and lower limit of the pixel after background model is trained, Y is the pixel in present frame
Gray value.
Further, it is described by the candidate target region comprising background by cluster screen after sort, remove false-alarm after
It is stored in the form of queue, when a new frame arrives, obtained candidate target region is put into queue and is owned with current queue
Target area matched, if successful match with current goal region replace queue in corresponding target area；If
It fails to match is then added to the target area end of queue, meanwhile, the target area duration of successful match in queue
Adding 1, the target area that it fails to match is deleted from queue, and countdown subtracts 1, when there are the target area that disappearance countdown is 0,
It is deleted from queue.
Further, the candidate target region matches with existing target area in current queue and should simultaneously meet such as
Lower requirement：
Each angular coordinate difference ＜ 20% in target area；
Pixel number number difference ＜ 20% in target area.
Further, the method that the target area after described pair of cluster screening is ranked up is as follows：
When the duration of the target area in object queue being more than 5, target area binding number 0-9 will be given, it is bound
Number cannot unbind automatically before the target area of binding is removed out queue, and each number can only bind a target simultaneously
Region has 10 target areas to be numbered every time.
This method of the present invention has the advantages that compared with prior art：The present invention, will by background modeling
Present image is matched with background image, and the binary map of candidate target region is obtained by analysis, the processing to matching result
Picture, by the way that binary image analysis, screening, sequencing numbers, multiple target under Condition of Complicated Ground Background is realized using the uniqueness of number
It quick and precisely detects, the method for the invention is simple and effective, it is easy to accomplish, solve that conventional method false-alarm is high, verification and measurement ratio is low, accidentally
High disadvantage is examined, the fast and effective detection to multiple target under static complex scene is realized.
Fast multi-target detection device under a kind of static complex scene of present invention proposition comprising the processor of interconnection
Module and editable device；Wherein, it is equipped with background modeling algorithm in editable device and code book compares algorithm, acquires input video
Acquisition input video in image, establish background model, by each pixel in current frame image background model carry out code
This comparison obtains background, the output of foreground bianry image and processor result；
Object filtering algorithm, number sorting algorithm are equipped in processor module, before comprising candidate target
Scape, background binary image removal false-alarm, the number for obtaining candidate target region and label are screened and are arranged to candidate target region
Sequence obtains the number and number in effective target region, and exports to editable logic gate module.
Processor module of the present invention is DSP module, and the editable device is FGPA modules.
The present invention simplifies optimization by being carried out to classical code book model in editable device, in combination in DSP module
To object filtering, sequence, the processing of number, solves the disadvantage that conventional method false-alarm is high, verification and measurement ratio is low, flase drop is high, realize
Fast and effective detection to multiple target under static complex scene achieves preferable effect in actual scene application.
Description of the drawings
Fig. 1 is the flow chart of the method for the invention.
Fig. 2 is the attachment structure schematic diagram of device of the present invention.
Fig. 3 is the flow chart of the method for the present invention background modeling method.
Fig. 4 is the flow chart that the method for the present invention is claim 6 matching process.
Fig. 5 is the present invention to the vehicle detection result figure outside 3 kilometers of target under static complex scene.
Fig. 6 is vehicle detection result figure of the present invention to multiple mobile object under haze weather under static complex scene.
Fig. 7 is the present invention to short distance, the vehicle result figure of target occlusion under static complex scene.
Specific implementation mode
It elaborates below in conjunction with the accompanying drawings to the embodiment of the present invention.The present embodiment before being with technical solution of the present invention
Put and implemented, give detailed embodiment and specific operating process, but protection scope of the present invention be not limited to
Under embodiment.
The present invention is based on fast multi-targets under a kind of static complex scene to detect embedded equipment, by taking vehicle detection as an example,
Input picture is the image sequence for including vehicle target under the static state complex scene of ground.
The present invention provides fast multi-targets under a kind of static complex scene to detect embedded equipment, shown in Fig. 2
FPGA module and the DSP module being connected with FPGA module realize detection, as shown in Figure 1, this method is as follows：
（1）, the 50-200 frame images that are acquired using FPGA module and preserve inputting video data, utilize background modeling algorithm to establish
Background model simultaneously stores；The quantity of image is selected according to the situation of change of scene and actual application demand；
（2）, come in present frame it is interim, by each pixel in image and the progress of established background model in FPGA module
Match, after successful match, which is labeled as background, is otherwise labeled as foreground, obtains the foreground and background for including candidate target
Bianry image, FPGA module are sent to DSP module by what is obtained comprising foreground and background bianry image；
（3）, DSP module to comprising foreground and background bianry image carry out clustering, obtain candidate target region number and
Label；
（4）, candidate target region mark is screened and is sorted, remove false-alarm, obtain the number and volume of true target area
Number, the numbered target area information of the tool of acquisition is sent to FPGA module by DSP module, is exported by FPGA module.
The present invention utilizes the realizability of FPGA module available resources and hardware, the performance of comprehensive assessment algorithm and in real time
Property.The present invention by the method for interpolation by the size reduction of original input picture to original 1/4, in this way, required storage
Amount reduces original 3/4, and operand also narrows down to original 1/4.
On the other hand, since classical code book model is directed to color video frequency image, R, G and B triple channel for using pixel are needed
Information, in order to further decrease storage and operand, the RGB image of input is converted YUV image by the present invention, and it is logical only to extract Y
The data in road are used for modeling, and compared with traditional rgb space, code word description is simpler, and amount of storage and operand greatly reduce.
The flow that background model is established using the codebook approach simplified after optimizing is as shown in Figure 3：
（1）In FPGA module, initialization code book model is set as sky, gray processing processing, input picture are carried out to input picture
Each pixel establish a code book, FPGA module reserves the space of 12 symbols to each pixel code book, and will have criterion
Will position is initialized as 0, and the study range of symbol determines that each code element structure is according to the reserved number of symbol space：
Ci={ Ymax, Ymin, Ylow, Yhigh, tlast, λ },
Wherein Ymax, YminThe respectively maximum and minimum value of current pixel gray value, Ylow, YhighFor symbol study lower limit and
The upper limit, initial value are the gray value of current pixel；tlastIt is renewal time, initial value 0 for symbol last time matching；λ
For symbol longest not renewal time, initial value 0, the value adds 1 when not updating；
（2）In modeling process, FPGA will be divided into four pieces per frame image, concurrent operation be carried out, when a new frame image data arrives
When coming, the code book time adds 1, and grey scale pixel value is limited to the value range of current grayvalue ［ ± 15, ± 25 ］, and the present invention is preferred
It is ± 20, therefore the study lower limit of code word is Ylow- 20, upper limit Yhigh+ 20, wherein code book timing definition is current modeling frame
Number, initial value 0, the modeling plus 1 per frame；
（3）During model training, if the gray value of some pixelLearn between the upper limit and lower limit in symbol, i.e. Ylow≤Y
≤Yhigh, then it is assumed that the pixel finds matched code word, then the code book time add 1, the maximin of more new symbol as the following formula, more
The study bound of the new symbol：
Ct={ max (Ymax, Y), min (Ymin, Y), Y-20, Y+20, t, λ }；
If code book is empty or matched symbol is not present, new symbol C is created as the following formulaL：
CL={ Ymax, Ymin, Ylow, Yhigh, 0,0 }；
Then the study bound of the symbol is updated, and its effective marker position is set as 1；
By other symbol longests of the code book, renewal time λ does not add 1；
At the end of modeling, check that all flag bits are 1 effective code element, if the longest of some symbol not renewal time λ ＞ 50,
Effective marker position is set as 0.
After completing background modeling, by the code book model of corresponding position pixel in each pixel of present frame and background model into
Row matching, matching rule are as follows：
Ylow≤Y≤Yhigh；
Wherein Ylow, YhighThe code book bound of the pixel is obtained for the background model training stage, Y is the pixel in present frame
Gray value.
Successful match is thought if meeting above formula, which is labeled as background, is denoted as 0, foreground is otherwise labeled as, is denoted as 1, obtains
The bianry image for including candidate target is obtained, obtained bianry image is sent to DSP module by FPGA module.
The present invention is needing to carry out 100 frame images respectively since the number of image frames used in background modeling is 100 frames
Modeling, after modeling, simplifies code book model.
As shown in figure 4, after model foundation, present frame that temporarily, the candidate target region that above-mentioned steps obtain is led to
It is ranked up after crossing cluster screening, is stored in the form of queue after removing false-alarm, when a new frame arrives, the candidate mesh that will obtain
Mark region is put into queue and is matched with target area all in current queue, and matching rule is as follows：Work as candidate target region
Following requirement should be met simultaneously by being matched with existing target area in current queue：Each angular coordinate difference ＜ 20% in target area；
In target area when pixel number number difference ＜ 20%, it is believed that successful match, otherwise for it fails to match.
It is replaced in queue and is corresponding to it with current goal region if successful match（Matching）Target area；If matching is lost
Lose, which be added to the end of queue, meanwhile, in queue the target area duration of successful match add 1,
With failure target area deleted from queue, countdown subtracts 1, when there are disappearance countdown be 0 target area when, by its from
It is deleted in queue.
Target area after being screened to cluster is ranked up, and ordering rule is as follows：When the target area in object queue
When duration is more than 5, number 0-9 will be bound to target area, bound number is removed out in the target area of binding
It cannot be unbinded automatically before queue, each number can only bind a target area simultaneously, have 10 target areas to obtain every time
Number.
In order to verify validity of the method for the present invention to multi-target detection in practical different complexity scenes, using actual field
Scape data are tested, and target is vehicle in scene.
Fig. 53 kilometers of targets under static complex scene are outer, the public affairs being located at 3 kilometers of the 1080p ball machines shooting on steel tower
The scene that road vehicle passes through, in the prior art, row's car that roadside is stopped have greatly the detection of actual motion vehicle
Interference, farther out due to distance, target imaging very little, while background clutter adds the interference of similar purpose, these are all given actually
Target detection made great challenge.And the method for the present invention can accurately exclude the interference of these pseudo- targets, it is final to obtain
Accurate testing result is very clear in red rectangle frame in figure.
Fig. 6 under haze weather, is located at high speed at 3 kilometers of the 1080p ball machines shooting on steel tower under static complex scene
Road scene, being interlocked and blocked due to haze influence, the interference of dynamic background, multiple target increases very greatly to actually detected task
Difficulty.The method of the present invention exclude dynamic background interference after, mark real target in figure in red rectangle frame very
Clearly, i.e., the vehicle moved on expressway.
Fig. 7 under static complex scene, network ball machine shooting 500m at urban road crossing scene, wherein target easily by
Trees, building block.From the results of view, the method for the present invention remains to realize under above-mentioned challenge the quick and precisely detection of target,
It is very clear in red rectangle frame in figure, largely solve the problems, such as in practical application.
In addition, in order to verify the advantage of the method for the present invention compared with prior art, using the static state of existing two kinds of mainstreams
Scene objects detection method GMM and VIBE is compared with the method for the present invention, using verification and measurement ratio, false alarm rate and computational complexity as
Evaluation criterion, in embedded platform（DSP+FPGA）On actual complex contextual data is tested, as a result as shown in table 1 below.
It is time-consuming higher since GMM is only emulated in PC machine, do not consider to realize on embedded platform.As can be seen from the table, with it is existing
Technology is compared, and the method for the present invention can obtain higher verification and measurement ratio and lower false alarm rate under complex scene, and by optimization
Algorithm runs real-time height on embedded platform afterwards, it is easy to accomplish.
1 the method for the present invention of table and prior art Contrast on effect：
。
The present invention simplifies optimization by being carried out to classical code book model in editable device, in combination in DSP module
To object filtering, sequence, the processing of number, solve that traditional multi-target detection method false-alarm is high, verification and measurement ratio is low, flase drop is high lacks
Point realizes the fast and effective detection to multiple target under static complex scene, and preferable effect is achieved in actual scene application
Fruit.
Part of that present invention that are not described in detail belong to the well-known technology of those skilled in the art.
Those of ordinary skill in the art it should be appreciated that more than embodiment be intended merely to illustrate the present invention,
And be not used as changing embodiment described above as long as in the spirit of the present invention for limitation of the invention,
Modification will all be fallen in the range of claims of the present invention.
Claims (10)
1. fast multi-target detection method under a kind of static state complex scene, which is characterized in that this method is as follows：
The image in input video is acquired, background model is established；
Each pixel in current frame image is matched with background model, carries out the label of background or foreground；
Foreground comprising candidate target, background binary image are handled, false-alarm is removed；
Candidate target region is screened and is sorted, the number and number in output effective target region；
Fast multi-target detection is completed under static complex scene.
2. fast multi-target detection method under static complex scene according to claim 1, which is characterized in that method is as follows：
At least 1 frame image for acquiring and preserving inputting video data, establishes background model using background modeling algorithm and stores；
Temporarily, each pixel in image to be matched with established background model, if successful match, by this in present frame
Pixel is labeled as background, is otherwise labeled as foreground；
The foreground and background bianry image for including candidate target is obtained, to carrying out cluster point comprising foreground and background bianry image
Analysis, obtains the number and label of candidate target region；
Above-mentioned candidate target region mark is screened and sorted, false-alarm is removed, obtains the number and number in real goal region
After export
Fast multi-target detection is completed under static complex scene.
3. fast multi-target detection method under static complex scene according to claim 2, which is characterized in that establish background mould
The method of type is as follows：
The image of modeling is obtained from inputting video data；
Several frame images are taken to be modeled；
Code book modeling is carried out to each pixel in each frame image, obtains code book model；
Code book model is simplified, all symbols in check image in each pixel code book；If some symbol in code book
Longest not renewal time λ ＞ 50, then be set as 0 by the corresponding flag bit of the symbol；
Complete background modeling.
4. fast multi-target detection method under static complex scene according to claim 3, which is characterized in that code book modeling
Method is as follows：
Initialization code book model is set as sky, gray processing processing is carried out to input picture, is built for each pixel of input picture
A code book is found, effective marker position is initialized as 0, the study range of symbol is determined according to the reserved number of symbol space, often
A symbol structures are Ci={ Ymax, Ymin, Ylow, Yhigh, tlast, λ },
Wherein Ymax, YminThe respectively maximum and minimum value of current pixel gray value, Ylow, YhighFor the study lower limit of symbol and upper
Limit, initial value is the gray value of current pixel；tlastIt is renewal time, initial value 0 for symbol last time matching；λ is
Symbol longest not renewal time, initial value 0, the value adds 1 when not updating；
It will be divided into four pieces per frame image, carry out concurrent operation, when a new frame image arrives, the code book time adds 1, by pixel
Gray value is limited to current grayvalue ± 20, therefore the study lower limit of code word is Ylow- 20, upper limit Yhigh+ 20, wherein code book
Timing definition is current modeling frame number, initial value 0, the modeling plus 1 per frame；
If the gray value of pixelLearn between the upper limit and lower limit in symbol, i.e. Ylow≤Y≤Yhigh, then successful match, when code book
Between plus 1, as the following formula the maximin of more new symbol, update the study bound of the symbol：
Ct={ max (Ymax, Y), min (Ymin, Y), Y-20, Y+20, t, λ }；
If the gray value Y of pixel is beyond between the symbol study upper limit and lower limit, then it is assumed that pixel does not find matched code word, then presses
Following formula creates new symbol CL, code book series adds 1：
CL={ Ymax, Ymin, Ylow, Yhigh, 0,0 }；
Then the study bound of the symbol is updated, and its effective marker position is set as 1；
By other symbol longests of the code book, renewal time λ does not add 1；
Check that all flag bits are 1 effective code element, if the longest of some symbol not renewal time λ ＞ 50, effective marker position
It is set as 0, modeling terminates.
5. fast multi-target detection method under static complex scene according to claim 4, which is characterized in that present frame it is every
A pixel is matched with the code book model of corresponding position pixel in background model, if the pixel is upper and lower in code book in present frame
Between limit, then it is denoted as background, is otherwise denoted as foreground；
Matching formula：Ylow≤Y≤Yhigh；
Wherein Ylow, YhighTo obtain the code book upper and lower limit of the pixel after background model is trained, Y is the pixel in present frame
Gray value.
6. fast multi-target detection method under static complex scene according to claim 5, which is characterized in that will include background
Candidate target region by cluster screen after sort, stored in the form of queue after removing false-alarm, will when a new frame arrives
Obtained candidate target region is put into queue and is matched with target area all in current queue, with working as if successful match
Replace corresponding target area in queue in preceding target area；The target area is added to the end of queue if it fails to match
End, meanwhile, in queue the target area duration of successful match add 1, the target area that it fails to match is deleted from queue,
Timing subtracts 1, and when there are the target area that disappearance countdown is 0, it is deleted from queue.
7. fast multi-target detection method under static complex scene according to claim 6, which is characterized in that candidate target area
Domain is matched with existing target area in current queue should meet following requirement simultaneously：
Each angular coordinate difference ＜ 20% in target area；
Pixel number number difference ＜ 20% in target area.
8. fast multi-target detection method under static complex scene according to claim 7, which is characterized in that screened to cluster
The method that target area afterwards is ranked up is as follows：
When the duration of the target area in object queue being more than 5, target area binding number 0-9 will be given, it is bound
Number cannot unbind automatically before the target area of binding is removed out queue, and each number can only bind a target simultaneously
Region has 10 target areas to be numbered every time.
9. the dress being detected using fast multi-target detection method under any one of the claim 1-8 static complex scenes
It sets, which is characterized in that the processor module including interconnection and editable device；
Wherein, it is equipped with background modeling algorithm in editable device and code book compares algorithm, the acquisition input for acquiring input video regards
Image in frequency establishes background model, each pixel in current frame image is carried out code book comparison in background model, is carried on the back
Scape, the output of foreground bianry image and processor result；
Object filtering algorithm, number sorting algorithm are equipped in processor module, before comprising candidate target
Scape, background binary image removal false-alarm, the number for obtaining candidate target region and label are screened and are arranged to candidate target region
Sequence obtains the number and number in effective target region, and exports to editable logic gate module.
10. fast multi-target detection device under static complex scene according to claim 9, which is characterized in that the processing
Device module is DSP module, and the editable device is FGPA modules.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810437311.2A CN108648210B (en) | 2018-05-09 | 2018-05-09 | Rapid multi-target detection method and device under static complex scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810437311.2A CN108648210B (en) | 2018-05-09 | 2018-05-09 | Rapid multi-target detection method and device under static complex scene |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108648210A true CN108648210A (en) | 2018-10-12 |
| CN108648210B CN108648210B (en) | 2022-06-14 |
Family
ID=63753937
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810437311.2A Active CN108648210B (en) | 2018-05-09 | 2018-05-09 | Rapid multi-target detection method and device under static complex scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108648210B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785356A (en) * | 2018-12-18 | 2019-05-21 | 北京中科晶上超媒体信息技术有限公司 | A kind of background modeling method of video image |
| CN112101135A (en) * | 2020-08-25 | 2020-12-18 | 普联国际有限公司 | Moving target detection method and device and terminal equipment |
| CN114694092A (en) * | 2022-03-15 | 2022-07-01 | 华南理工大学 | Expressway monitoring video object-throwing detection method based on mixed background model |
| CN116228544A (en) * | 2023-03-15 | 2023-06-06 | 阿里巴巴（中国）有限公司 | Image processing method and device and computer equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103325259A (en) * | 2013-07-09 | 2013-09-25 | 西安电子科技大学 | Illegal parking detection method based on multi-core synchronization |
| US20140369552A1 (en) * | 2013-06-14 | 2014-12-18 | National Yunlin University Of Science And Technology | Method of Establishing Adjustable-Block Background Model for Detecting Real-Time Image Object |
| CN104835163A (en) * | 2015-05-11 | 2015-08-12 | 华中科技大学 | Embedded real-time high-speed binocular vision system for moving target detection |
| CN105389831A (en) * | 2015-11-11 | 2016-03-09 | 南京邮电大学 | Multi-target detection method based on YUV color space |
-
2018
- 2018-05-09 CN CN201810437311.2A patent/CN108648210B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140369552A1 (en) * | 2013-06-14 | 2014-12-18 | National Yunlin University Of Science And Technology | Method of Establishing Adjustable-Block Background Model for Detecting Real-Time Image Object |
| CN103325259A (en) * | 2013-07-09 | 2013-09-25 | 西安电子科技大学 | Illegal parking detection method based on multi-core synchronization |
| CN104835163A (en) * | 2015-05-11 | 2015-08-12 | 华中科技大学 | Embedded real-time high-speed binocular vision system for moving target detection |
| CN105389831A (en) * | 2015-11-11 | 2016-03-09 | 南京邮电大学 | Multi-target detection method based on YUV color space |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109785356A (en) * | 2018-12-18 | 2019-05-21 | 北京中科晶上超媒体信息技术有限公司 | A kind of background modeling method of video image |
| CN109785356B (en) * | 2018-12-18 | 2021-02-05 | 北京中科晶上超媒体信息技术有限公司 | Background modeling method for video image |
| CN112101135A (en) * | 2020-08-25 | 2020-12-18 | 普联国际有限公司 | Moving target detection method and device and terminal equipment |
| CN114694092A (en) * | 2022-03-15 | 2022-07-01 | 华南理工大学 | Expressway monitoring video object-throwing detection method based on mixed background model |
| CN116228544A (en) * | 2023-03-15 | 2023-06-06 | 阿里巴巴（中国）有限公司 | Image processing method and device and computer equipment |
| CN116228544B (en) * | 2023-03-15 | 2024-04-26 | 阿里巴巴（中国）有限公司 | Image processing method and device and computer equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108648210B (en) | 2022-06-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108648210A (en) | It is a kind of static state complex scene under fast multi-target detection method and device | |
| CN109903312A (en) | A kind of football sportsman based on video multi-target tracking runs distance statistics method | |
| CN110136154A (en) | Remote sensing images semantic segmentation method based on full convolutional network and Morphological scale-space | |
| CN111784685A (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN108288075A (en) | A kind of lightweight small target detecting method improving SSD | |
| CN105975929A (en) | Fast pedestrian detection method based on aggregated channel features | |
| CN111985499B (en) | High-precision bridge apparent disease identification method based on computer vision | |
| CN105574550A (en) | Vehicle identification method and device | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| CN113658192B (en) | Multi-target pedestrian track acquisition method, system, device and medium | |
| CN106778734A (en) | A kind of insulator based on rarefaction representation falls to go here and there defect inspection method | |
| CN113128335B (en) | Method, system and application for detecting, classifying and finding micro-living ancient fossil image | |
| CN106157323A (en) | The insulator division and extracting method that a kind of dynamic division threshold value and block search combine | |
| CN108664939A (en) | A kind of remote sensing images aircraft recognition method based on HOG features and deep learning | |
| CN110222604A (en) | Target identification method and device based on shared convolutional neural networks | |
| CN111967313A (en) | Unmanned aerial vehicle image annotation method assisted by deep learning target detection algorithm | |
| CN113435407B (en) | Small target identification method and device for power transmission system | |
| CN110059539A (en) | A kind of natural scene text position detection method based on image segmentation | |
| CN112489055B (en) | Satellite video dynamic vehicle target extraction method fusing brightness-time sequence characteristics | |
| CN106503638A (en) | For the image procossing of colour recognition, vehicle color identification method and system | |
| CN109815798A (en) | Unmanned plane image processing method and system | |
| Zhang et al. | Application research of YOLO v2 combined with color identification | |
| CN109785288A (en) | Transmission facility defect inspection method and system based on deep learning | |
| CN110516707A (en) | A kind of image labeling method and its device, storage medium | |
| CN110135248A (en) | A kind of natural scene Method for text detection based on deep learning |
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 |