CN108416283A - A kind of pavement marking recognition methods based on SSD - Google Patents
A kind of pavement marking recognition methods based on SSD Download PDFInfo
- Publication number
- CN108416283A CN108416283A CN201810168081.4A CN201810168081A CN108416283A CN 108416283 A CN108416283 A CN 108416283A CN 201810168081 A CN201810168081 A CN 201810168081A CN 108416283 A CN108416283 A CN 108416283A
- Authority
- CN
- China
- Prior art keywords
- pavement marking
- layers
- recognition methods
- ssd
- layer
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The pavement marking recognition methods based on SSD that the invention discloses a kind of, including step:Carry out Image Acquisition and pretreatment, makes sample collection;Training set is inputted, SSD networks are trained in multitask;Picture to be identified obtains characteristic pattern by several convolutional layers and pond layer;The characteristic pattern of the convolutional layer output different to wherein 5 kinds carries out convolution with two 3 × 3 different convolution kernels respectively, respectively obtains classification score and window returns two output vectors;All results are handled by non-maxima suppression and generate final Target detection and identification as a result, traffic sign is identified.The present invention is using this deep learning method of SSD, compared with Faster R CNN, the process of candidate frame is not generated, realize multitask training, additional characteristic storage space is not needed yet, detection speed and precision are improved, compared to shallow-layer Study strategies and methods, it has higher learning efficiency and accuracy of identification.
Description
Technical field
The invention belongs to image procossings and automotive safety auxiliary driving field, more particularly, to a kind of road surface based on SSD
Road traffic sign detection and recognition methods, to solve the problems, such as accuracy of identification is not high in pavement marking identification problem.
Background technology
Traffic sign recognition (TSR, Traffic Signs Recognition) is as one in vehicle-mounted auxiliary system
Important branch is one of current still unsolved problem.Due in traffic sign containing there are many important traffic information, such as to working as
How soon the speed prompt of preceding driving, the variation of road ahead situation, driving behavior restrict, therefore in the auxiliary system,
Speed accurately and efficiently identifies the traffic sign in road and it is fed back to driver or control system, and guarantee is driven
Safety is sailed, the generation to avoid traffic accident has highly important research significance.
Pavement marking identifies that common method includes the recognition methods based on shape, and feature extraction and classifying device combines
Method, the recognition methods of deep learning.Recognition methods robustness based on shape is poor, ineffective in complex environment.It is special
The method recognition effect that sign extraction is combined with grader is preferable, but computing cost is big, and adaptive capacity to environment is poor.Deep learning
Directly original image can be identified, the recessive character of extraction reflection data essence has enough study depth.Convolution
Situations such as characteristic that neural network is shared with local weight, complicated, multi-angle changes for environment, all has centainly real-time
Property and robustness.Therefore, it is necessary to design a kind of recognition methods that can accurately obtain pavement marking in road scene.Wei
The SSD algorithms that Liu was proposed in 2016, have evaded Faster R-CNN (FasterRegion-based Convolutional
Neural Network) in redundancy feature extraction operation, realize multitask training, it is empty not need additional characteristic storage yet
Between, improve detection speed and precision.
Invention content
To solve the above problem of the existing technology, the present invention will design a kind of based on the identification of SSD pavement markings
Method can accurately obtain pavement marking in road scene, to help to assist driver under conditions of complexity
Vehicle external environment is preferably perceived, traffic accident is prevented.
To achieve the goals above, technical scheme is as follows:
A kind of pavement marking recognition methods based on SSD, including step:
Carry out Image Acquisition and pretreatment, makes sample collection;
Training set is inputted, starts multitask and trains SSD networks;
Picture input SSD networks to be trained obtain characteristic pattern by several convolutional layers and pond layer;
A series of acquiescence frame of fixed sizes is generated on the characteristic pattern of 5 different layers wherein, and therefrom generates positive and negative sample
This collection ensures its ratio 1:3;
The feature vector of extraction respectively obtains classification score and window returns two outputs via multilayer fusion detection network
Vector;
After the completion of network training, picture to be identified is inputted into trained network, all results are subjected to non-maximum suppression
System processing generates final Target detection and identification as a result, to enable pavement marking to identify.
Further, the step of described image acquires specifically includes:
Open vehicle-mounted traveling recorder, captured in real-time road traffic video information;
The video information that video camera is taken carries out sub-frame processing, obtains an image collection sequence;
Image collection is screened, the image for including pavement marking is chosen.
Further, the step of described image pretreatment and sample set make specifically includes:
In the image of selection, target area is taken out, and zooms to fixed 300 × 300 size, is compared for enhancing
Degree, then contrast enhancement processing is passed through into target area, original training set is obtained, test set is handled in the same way;
It takes out in data set and collects with the comparable sample composition verification of test set number at random, remaining sample composition is final
Training set.
Further, the SSD network structures include:Data Layer, feature extraction network, multilayer fusion detection network,
Concat layers and loss layer.
The feature extraction network is mainly made of two parts, including VGGNet and 8 newly-increased convolutional layers, described
VGGNet was changed, complete entitled VGG_ILSVRC_16_layers_fc_reduced, and newly-increased 8 convolutional layers are used for
The layer for generating more low resolution facilitates and carries out multilayer feature fusion.
Further, the multilayer fusion detection network includes loc layers, conf layers and priorbox layers.
Further, Concat layers of the quantity is three, for splicing all loc layers respectively, conf layers and
Priorbox layers.
Further, a series of acquiescence frame of fixed sizes is generated on the characteristic pattern of 5 different layers wherein, and from
It is middle to generate positive and negative sample set, ensure its ratio 1:3 the step of, specifically includes:
A series of acquiescence frame (default box) of fixed sizes is generated on the characteristic pattern of 5 different layers wherein, therefrom
The conduct candidate's positive sample collection for meeting true frame (groundtruth box) is selected, remaining is then candidate negative sample collection;Further
Screening Samples obtain positive and negative sample set, ensure that the ratio of positive sample collection and negative sample collection is 1:3.
Further, the loss function of the SSD networks includes Liang Ge branches, be respectively used to classify and position, using with
Machine gradient descent method trains the classification layer of output layer and returns layer until classification and the loss function convergence returned.
Further, described the step of all results are carried out non-maxima suppression processing, specifically includes:According to output
Liang Ge branches carry out non-maxima suppression to each type objects respectively using window score and reject overlapping frame, finally obtain each
The window of revised highest scoring is returned in classification.
Further, the pavement marking includes straight trip arrow, the arrow that turns around, to the left arrow, right-hand arrow, straight trip
Arrow, straight trip right-hand arrow and diamond shape graticule to the left.
Compared with prior art, the present invention is above-mentioned in the prior art at least some in order to solve the problems, such as, it is proposed that Yi Zhongji
Road traffic sign detection in SSD and recognition methods.This method self manufacture pavement marking data set, by deep learning from
Learning characteristic in sample can extract the recessive character of reflection data essence, have higher learning efficiency and accuracy of identification,
The robustness for improving detection algorithm effectively increases the accuracy of pavement marking detection.
Description of the drawings
The present invention provides attached drawing further understanding in order to disclosure, and attached drawing constitutes part of this application,
But it is only used for illustrating the non-limiting example for some inventions for embodying concept of the invention, rather than for making any limit
System.
Fig. 1 is the flow of the pavement marking recognition methods based on SSD of some example embodiments according to the present invention
Figure.
Fig. 2 is the SSD network structures of some example embodiments according to the present invention.
Fig. 3 is the multitask training loss function schematic diagram of some example embodiments according to the present invention.
Fig. 4 is the schematic diagram of the part traffic sign sample set of some example embodiments according to the present invention.
Fig. 5 is the detection of the pavement marking recognition methods based on SSD of some example embodiments according to the present invention
Result schematic diagram.
Specific implementation mode
It elaborates below in conjunction with the accompanying drawings to the present invention with technical solution.
It is the flow chart of the pavement marking recognition methods based on SSD, specific embodiment party of the invention as shown in Fig. 1
Formula is:
A kind of pavement marking recognition methods based on SSD, including step:
Carry out Image Acquisition and pretreatment, makes sample collection;
Training set is inputted, starts multitask and trains SSD networks;
Picture input SSD networks to be trained obtain characteristic pattern by several convolutional layers and pond layer;
A series of acquiescence frame of fixed sizes is generated on the characteristic pattern of 5 different layers wherein, and therefrom generates positive and negative sample
This collection ensures its ratio 1:3;
The feature vector of extraction respectively obtains classification score and window returns two outputs via multilayer fusion detection network
Vector;
After the completion of network training, picture to be identified is inputted into trained network, all results are subjected to non-maximum suppression
System processing generates final Target detection and identification as a result, to enable pavement marking to identify.
In feasible embodiment, described image acquire the step of specifically include:
Vehicle-mounted traveling recorder is opened, captured in real-time road traffic video information selects the video of automobile data recorder shooting
Resolution ratio is the video image of 1280*720;
Sub-frame processing is carried out to captured video image, obtains an image collection sequence;
Image collection is screened, the 7 kind pavement markings more from wherein selection occurrence number.
Specifically, described image pretreatment and sample set specifically include the step of making:
In the image of selection, target area is taken out, and zooms to fixed 300 × 300 size, is compared for enhancing
Degree, then contrast enhancement processing is passed through into target area, original training set is constituted, test set is handled in the same way;
It takes out in data set and collects with the comparable sample composition verification of test set number at random, remaining sample composition is final
Training set.
In the example depicted in fig. 4, selected traffic sign can be divided into 7 classes, and arrow of respectively keeping straight on, turn around arrow
Head, to the left arrow, right-hand arrow, keep straight on arrow, straight trip right-hand arrow, diamond shape graticule to the left, number be expressed as 01,02 respectively,
03,04,05,06,07, recognition result exports by this method.
K.Simonyan et al. is in document " K.Simonyan and A.Zisserman.Very deep
Convolutional networks for large-scale image recognition, 2015. " the middle VGG16 proposed
Network, including 13 convolutional layers, 5 pond layers and 3 full articulamentums.The present embodiment is modified on the basis of VGG16, is obtained
To new VGG_ILSVRC_16_layers_fc_reduced, and 8 convolutional layers are increased newly, to generate more low resolution
Layer facilitates and carries out multilayer feature fusion.
The multilayer fusion detection network, including loc layers, conf layers and priorbox layers.
Concat layers of the quantity is three, splices all loc layers respectively, conf layers and priorbox layers.
As shown in Fig. 2, it is as follows to finally obtain SSD network structures:Including data Layer, feature extraction network, multilayer fusion inspection
Survey grid network, Concat layers and loss layer.
The image pattern that size is 300 × 300 inputs network through input layer;Conv4_3, conv7 (original FC7),
Conv8_2, conv9_2, conv10_2 and pool11, these layers are predicted to return position and confidence level.Activation primitive uses
Linear unit activating (ReLU, Rectified Linear Units) function is corrected, has the sparse ability of guiding appropriateness, it can
The training speed of network is set to accelerate, precision improves, and avoids the problem that gradient disappears.
Original layer parameter will be initialized by pre-training mode.For classification full articulamentum with mean value be 0, standard deviation
It is initialized for 0.01 Gaussian Profile;Full articulamentum for recurrence is with mean value for 0, and standard deviation is at the beginning of 0.001 Gaussian Profile
Beginningization, biasing are initialized to 0.
SSD networks carry out multitask training, and confidence level loss and recurrence loss are as shown in Figure 3.
It allows each acquiescence frame to be calculated by Jaccard coefficients and the similarity of true frame, threshold value only has the ability more than 0.5
Short-list can be classified as;Assuming that is chosen is the frame that N number of matching degree is higher than 50 percent, we enable i indicate i-th
Give tacit consent to frame, j indicates that j-th of true frame, p indicate p-th of class.So xij pIndicate j-th of i-th acquiescence frame and classification p it is true
The Jaccard coefficients that frame matches, if if mismatching, xij p=0.
Jaccard coefficients are used for comparing similarity between two finite aggregates, and bigger correlation is higher.
Confidence level loses Lconf(confidence loss) is Softmax Loss, is inputted as the confidence level c of every one kind.
Positioning loss Lloc(localization loss) is Smooth L1Loss is used in prediction block (l) and true frame
(g) parameter (i.e. centre coordinate position (cx, cy), in width w, height h), return bounding boxes center and
Width and height.
Combining classification loses and returns loss, and the network fine tuning stage, total loss function was:
Wherein, N is the acquiescence frame number to match with true frame.Weight term α default settings are 1.According to the loss function
Network is trained using stochastic gradient descent method, until L restrains.
Fig. 5 is the testing result schematic diagram of pavement marking, it is seen that effect is detected and identified under general pavement conditions
Fruit is good.
To sum up, the present invention proposes a kind of road traffic sign detection based on SSD and recognition methods.This method passes through depth
The learning characteristic from sample is practised, the recessive character of reflection data essence can be extracted, there is higher learning efficiency and identification
Precision improves the robustness of detection algorithm, effectively increases the accuracy of pavement marking detection.Can largely it delay
Solution pavement marking serious shielding, the detection difficult that factors bring such as serious wear, deformation is serious, illumination variation is serious are asked
Topic.Part Methods step herein and flow may need to be executed by computer, to hardware, software, firmware and its appoint
The mode of what combination is implemented.
The preferred embodiment of the present invention is above are only, implementation and the interest field of invention are not intended to limit, it is all according to this
Equivalence changes, modification, replacement that content described in patent application scope of patent protection is made etc. should all be included in the present patent application
In the scope of the claims.Those skilled in the art will appreciate that without departing from the scope and spirit of the present invention, it can be wider
It is changed and modified in wealthy various aspects.
Claims (10)
1. a kind of pavement marking recognition methods based on SSD, which is characterized in that including step:
Carry out Image Acquisition and pretreatment, makes sample collection;
Training set is inputted, starts multitask and trains SSD networks;
Picture input SSD networks to be trained obtain characteristic pattern by several convolutional layers and pond layer;
A series of acquiescence frame of fixed sizes is generated on the characteristic pattern of 5 different layers wherein, and therefrom generates positive and negative sample set,
Ensure its ratio 1:3;
By the feature vector of extraction via multilayer fusion detection network, respectively obtain classification score and window return two export to
Amount;
After the completion of network training, picture to be identified is inputted into trained network, all results are carried out at non-maxima suppression
Reason generates final Target detection and identification as a result, to enable pavement marking to identify.
2. pavement marking recognition methods as described in claim 1, it is characterised in that:The step of described image acquires is specific
Including:
Open vehicle-mounted traveling recorder, captured in real-time road traffic video information;
The video information that video camera is taken carries out sub-frame processing, obtains an image collection sequence;
Image collection is screened, the image for including pavement marking is chosen.
3. pavement marking recognition methods as described in claim 1, it is characterised in that:Described image pre-processes and sample set
The step of making, specifically includes:
In the image of selection, target area is taken out, and zooms to fixed 300 × 300 size, to enhance contrast, then
Contrast enhancement processing is passed through into target area, obtains original training set, test set is handled in the same way;
It is taken out at random in data set and forms final instruction with the comparable sample composition verification collection of test set number, remaining sample
Practice collection.
4. pavement marking recognition methods as described in claim 1, it is characterised in that:The SSD network structures include:Number
According to layer, feature extraction network, multilayer fusion detection network, Concat layers and loss layer.
The feature extraction network is mainly made of two parts, including VGGNet and 8 newly-increased convolutional layers, described 8 newly-increased
Convolutional layer is used to generate the layer of more low resolution, facilitates and carries out multilayer feature fusion.
5. pavement marking recognition methods as claimed in claim 4, it is characterised in that:The multilayer fusion detection network
Including loc layers, conf layers and priorbox layers.
6. pavement marking recognition methods as claimed in claim 5, it is characterised in that:Concat layers of the quantity is three
It is a, for splicing all loc layers respectively, conf layers and priorbox layers.
7. pavement marking recognition methods as described in claim 1, it is characterised in that:5 different layers wherein
A series of acquiescence frame of fixed sizes is generated on characteristic pattern, and therefrom generates positive and negative sample set, ensures its ratio 1:3 the step of
It specifically includes:
A series of acquiescence frame for generating fixed sizes on the characteristic pattern of 5 different layers wherein, therefrom selects and meets true frame
As candidate positive sample collection, remaining is then candidate negative sample collection;Further Screening Samples obtain positive and negative sample set, ensure positive sample
The ratio of collection and negative sample collection is 1:3.
8. pavement marking recognition methods as described in claim 1, it is characterised in that:The loss function of the SSD networks
It including Liang Ge branches, is respectively used to classify and position, train the classification layer of output layer using stochastic gradient descent method and return layer
Until classification and the loss function convergence returned.
9. pavement marking recognition methods as described in claim 1, it is characterised in that:It is described that all results are subjected to non-pole
The step of big value inhibits processing specifically includes:According to the Liang Ge branches of output, using window score respectively to each type objects into
Row non-maxima suppression rejects overlapping frame, finally obtains the window that revised highest scoring is returned in each classification.
10. pavement marking recognition methods as claimed in any one of claims 1-9 wherein, it is characterised in that:It hands on the road surface
Logical mark includes straight trip arrow, the arrow that turns around, arrow, right-hand arrow, straight trip to the left arrow, straight trip right-hand arrow and diamond shape to the left
Graticule.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810168081.4A CN108416283A (en) | 2018-02-28 | 2018-02-28 | A kind of pavement marking recognition methods based on SSD |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810168081.4A CN108416283A (en) | 2018-02-28 | 2018-02-28 | A kind of pavement marking recognition methods based on SSD |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108416283A true CN108416283A (en) | 2018-08-17 |
Family
ID=63129424
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810168081.4A Pending CN108416283A (en) | 2018-02-28 | 2018-02-28 | A kind of pavement marking recognition methods based on SSD |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108416283A (en) |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109325438A (en) * | 2018-09-18 | 2019-02-12 | 桂林电子科技大学 | The real-time identification method of live panorama traffic sign |
CN109359553A (en) * | 2018-09-21 | 2019-02-19 | 上海小萌科技有限公司 | Commodity detection method, device, computer equipment and the storage medium of fish eye images |
CN109660761A (en) * | 2018-12-21 | 2019-04-19 | 鞍钢集团自动化有限公司 | The personal identification of band conveyer corridor complex environment and alarm system and method based on SSD |
CN109741318A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | The real-time detection method of single phase multiple dimensioned specific objective based on effective receptive field |
CN109753949A (en) * | 2019-01-31 | 2019-05-14 | 南京邮电大学 | A kind of more form method for traffic sign detection based on deep learning |
CN109753959A (en) * | 2018-12-21 | 2019-05-14 | 西北工业大学 | Road traffic sign detection method based on self-adaptive multi-scale feature fusion |
CN109977942A (en) * | 2019-02-02 | 2019-07-05 | 浙江工业大学 | A kind of scene character recognition method based on scene classification and super-resolution |
CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
CN111241916A (en) * | 2019-12-25 | 2020-06-05 | 苏州奥易克斯汽车电子有限公司 | Method for establishing traffic sign recognition model |
CN111401466A (en) * | 2020-03-26 | 2020-07-10 | 广州紫为云科技有限公司 | Traffic sign detection and identification marking method and device and computer equipment |
CN111985373A (en) * | 2020-08-12 | 2020-11-24 | 杭州鸿泉物联网技术股份有限公司 | Safety early warning method and device based on traffic intersection identification and electronic equipment |
CN112149624A (en) * | 2020-10-16 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Traffic identification image processing method and device |
CN112528971A (en) * | 2021-02-07 | 2021-03-19 | 北京智芯微电子科技有限公司 | Power transmission line abnormal target detection method and system based on deep learning |
CN112712066A (en) * | 2021-01-19 | 2021-04-27 | 腾讯科技(深圳)有限公司 | Image recognition method and device, computer equipment and storage medium |
CN113076800A (en) * | 2021-03-03 | 2021-07-06 | 惠州市博实结科技有限公司 | Road sign board detection method and device |
CN113160586A (en) * | 2021-03-26 | 2021-07-23 | 深圳职业技术学院 | Tide channel control method, device and system thereof, and computer readable storage medium |
CN113591543A (en) * | 2021-06-08 | 2021-11-02 | 广西综合交通大数据研究院 | Traffic sign recognition method and device, electronic equipment and computer storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140247987A1 (en) * | 2013-03-04 | 2014-09-04 | Megachips Corporation | Object detection apparatus, storage medium, and integrated circuit |
CN107301383A (en) * | 2017-06-07 | 2017-10-27 | 华南理工大学 | A kind of pavement marking recognition methods based on Fast R CNN |
CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技(深圳)有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
CN107704836A (en) * | 2017-10-17 | 2018-02-16 | 电子科技大学 | Method for detecting fatigue driving based on object detection |
-
2018
- 2018-02-28 CN CN201810168081.4A patent/CN108416283A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20140247987A1 (en) * | 2013-03-04 | 2014-09-04 | Megachips Corporation | Object detection apparatus, storage medium, and integrated circuit |
CN107301383A (en) * | 2017-06-07 | 2017-10-27 | 华南理工大学 | A kind of pavement marking recognition methods based on Fast R CNN |
CN107679531A (en) * | 2017-06-23 | 2018-02-09 | 平安科技(深圳)有限公司 | Licence plate recognition method, device, equipment and storage medium based on deep learning |
CN107704836A (en) * | 2017-10-17 | 2018-02-16 | 电子科技大学 | Method for detecting fatigue driving based on object detection |
Cited By (24)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109325438B (en) * | 2018-09-18 | 2021-06-15 | 桂林电子科技大学 | Real-time identification method of live panoramic traffic sign |
CN109325438A (en) * | 2018-09-18 | 2019-02-12 | 桂林电子科技大学 | The real-time identification method of live panorama traffic sign |
CN109359553A (en) * | 2018-09-21 | 2019-02-19 | 上海小萌科技有限公司 | Commodity detection method, device, computer equipment and the storage medium of fish eye images |
CN109660761A (en) * | 2018-12-21 | 2019-04-19 | 鞍钢集团自动化有限公司 | The personal identification of band conveyer corridor complex environment and alarm system and method based on SSD |
CN109753959A (en) * | 2018-12-21 | 2019-05-14 | 西北工业大学 | Road traffic sign detection method based on self-adaptive multi-scale feature fusion |
CN109753959B (en) * | 2018-12-21 | 2022-05-13 | 西北工业大学 | Road traffic sign detection method based on self-adaptive multi-scale feature fusion |
CN109741318A (en) * | 2018-12-30 | 2019-05-10 | 北京工业大学 | The real-time detection method of single phase multiple dimensioned specific objective based on effective receptive field |
CN109741318B (en) * | 2018-12-30 | 2022-03-29 | 北京工业大学 | Real-time detection method of single-stage multi-scale specific target based on effective receptive field |
CN109753949A (en) * | 2019-01-31 | 2019-05-14 | 南京邮电大学 | A kind of more form method for traffic sign detection based on deep learning |
CN109753949B (en) * | 2019-01-31 | 2022-08-12 | 南京邮电大学 | Multi-window traffic sign detection method based on deep learning |
CN109977942B (en) * | 2019-02-02 | 2021-07-23 | 浙江工业大学 | Scene character recognition method based on scene classification and super-resolution |
CN109977942A (en) * | 2019-02-02 | 2019-07-05 | 浙江工业大学 | A kind of scene character recognition method based on scene classification and super-resolution |
CN110472503A (en) * | 2019-07-11 | 2019-11-19 | 桂林电子科技大学 | A kind of road target detection method, device and storage medium |
CN111241916A (en) * | 2019-12-25 | 2020-06-05 | 苏州奥易克斯汽车电子有限公司 | Method for establishing traffic sign recognition model |
CN111401466A (en) * | 2020-03-26 | 2020-07-10 | 广州紫为云科技有限公司 | Traffic sign detection and identification marking method and device and computer equipment |
CN111985373A (en) * | 2020-08-12 | 2020-11-24 | 杭州鸿泉物联网技术股份有限公司 | Safety early warning method and device based on traffic intersection identification and electronic equipment |
CN112149624A (en) * | 2020-10-16 | 2020-12-29 | 腾讯科技(深圳)有限公司 | Traffic identification image processing method and device |
CN112712066A (en) * | 2021-01-19 | 2021-04-27 | 腾讯科技(深圳)有限公司 | Image recognition method and device, computer equipment and storage medium |
CN112528971B (en) * | 2021-02-07 | 2021-06-04 | 北京智芯微电子科技有限公司 | Power transmission line abnormal target detection method and system based on deep learning |
CN112528971A (en) * | 2021-02-07 | 2021-03-19 | 北京智芯微电子科技有限公司 | Power transmission line abnormal target detection method and system based on deep learning |
CN113076800A (en) * | 2021-03-03 | 2021-07-06 | 惠州市博实结科技有限公司 | Road sign board detection method and device |
CN113160586A (en) * | 2021-03-26 | 2021-07-23 | 深圳职业技术学院 | Tide channel control method, device and system thereof, and computer readable storage medium |
CN113591543A (en) * | 2021-06-08 | 2021-11-02 | 广西综合交通大数据研究院 | Traffic sign recognition method and device, electronic equipment and computer storage medium |
CN113591543B (en) * | 2021-06-08 | 2024-03-26 | 广西综合交通大数据研究院 | Traffic sign recognition method, device, electronic equipment and computer storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108416283A (en) | A kind of pavement marking recognition methods based on SSD | |
Wang et al. | Weakly supervised adversarial domain adaptation for semantic segmentation in urban scenes | |
CN107301383A (en) | A kind of pavement marking recognition methods based on Fast R CNN | |
CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
Pereira et al. | A deep learning-based approach for road pothole detection in timor leste | |
Ahmad et al. | Small object detection in aerial imagery using RetinaNet with anchor optimization | |
CN103761747A (en) | Target tracking method based on weighted distribution field | |
Guan et al. | Multi-scale object detection with feature fusion and region objectness network | |
Asgarian Dehkordi et al. | Vehicle type recognition based on dimension estimation and bag of word classification | |
Anwer et al. | Accident vehicle types classification: a comparative study between different deep learning models | |
Sun et al. | IRDCLNet: Instance segmentation of ship images based on interference reduction and dynamic contour learning in foggy scenes | |
CN112347967B (en) | Pedestrian detection method fusing motion information in complex scene | |
Zhang et al. | A quality index metric and method for online self-assessment of autonomous vehicles sensory perception | |
Baruah et al. | Analysis of Traffic Sign Recognition for Automated Transportation Systems Using Neural Networks | |
CN115761667A (en) | Unmanned vehicle carried camera target detection method based on improved FCOS algorithm | |
Filatov et al. | Application of machine learning methods for traffic signs recognition | |
Kinattukara et al. | Clustering based neural network approach for classification of road images | |
Xiong et al. | Vehicle detection algorithm based on lightweight YOLOX | |
Xue et al. | Vehicle detection and pose estimation by probabilistic representation | |
Zhao et al. | Leveraging Monte Carlo Dropout for Uncertainty Quantification in Real-Time Object Detection of Autonomous Vehicles | |
Shantaiya et al. | Multiple class image-based vehicle classification using soft computing algorithms. | |
Dai et al. | An adaptive sample assignment network for tiny object detection | |
Wang et al. | YOLOX-based dynamic vehicle target detection | |
Ashik et al. | Recognizing bangladeshi traffic signs in the wild | |
Zhang et al. | Novel freight train image fault detection and classification models based on CNN |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180817 |
|
RJ01 | Rejection of invention patent application after publication |