CN109508635A - A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network - Google Patents
A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network Download PDFInfo
- Publication number
- CN109508635A CN109508635A CN201811167453.8A CN201811167453A CN109508635A CN 109508635 A CN109508635 A CN 109508635A CN 201811167453 A CN201811167453 A CN 201811167453A CN 109508635 A CN109508635 A CN 109508635A
- Authority
- CN
- China
- Prior art keywords
- label
- tensorflow
- network
- traffic
- picture
- 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
- 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
- G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network belongs to computer vision and machine learning field;Prepare traffic signals identification public data collection image, label and sets of video data;Using OpenCV to normalization numerical value TensorFlow and traffic signals identification public data the image modification size concentrated and export tri- channels its RGB;Convert simultaneously normalized according to the transformation rule of figure to coordinate label;The picture containing traffic lights information is extracted according to tab indexes and corresponds it with label;Picture and label are sent into CNN network and are trained simultaneously preservation model;Keep the accuracy of effective solution of the present invention mark not high and the slow technical problem of recognition speed.
Description
Technical field
The invention belongs to computer visions and machine learning field, more particularly to one kind to be based on TensorFlow combination multilayer
The traffic light recognition method of CNN network.
Background technique
Carrying out identification to the traffic lights position in image is a combination computer vision and machine learning (Machine
Learning technology), with the arrival in artificial intelligence epoch, the receipts of equipment based on autonomous driving vehicle to traffic information
The needs for collecting processing are increasing, how to find accurate traffic information in the range more efficiently shown in camera, only
It is inadequate efficiently and accurately by traditional image recognition algorithm.For autonomous driving vehicle, in order to guarantee existing
It drives safely under laws and regulations and traffic rules, the real-time perception ability to road conditions is highly important, wherein the important point
It is exactly the collection processing capacity of traffic lights information, since traffic lights information change is very fast, it is desirable that automated driving system is in this respect
Collection processing want speed fast, accuracy is high.
The mark to traffic lights position in picture also may be implemented by traditional images processing means, but its mark is accurate
It is slow to spend not high and recognition speed.
Summary of the invention
The present invention overcomes above-mentioned the deficiencies in the prior art, provide a kind of based on TensorFlow combination multi-layer C NN network
Traffic light recognition method, the present invention using traffic signals identify public data collection, i.e. Traffic Lights
Recognition (TLR) public benchmarks data set the most contains the effective picture of 6228 width and corresponding one-to-many
Label, the characteristics of image for extracting the present invention has certain expressive faculty, by convolutional neural networks algorithm to picture
In traffic lights position be labeled, make that effective solution of the present invention marks accuracy is not high and the slow technology of recognition speed
Problem.
Technical solution of the present invention:
A kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network, comprising the following steps:
Step a, prepare traffic signals identification public data collection image, label and sets of video data;
Step b, to the image modification size of traffic signals identification public data concentration and its RGB tri- is exported with OpenCV
The normalization numerical value in channel;
Step c, convert simultaneously normalized according to the transformation rule of figure to coordinate label;
Step d, the picture containing traffic lights information is extracted according to tab indexes and correspond it with label;
Step e, picture and label are sent into CNN network and are trained simultaneously preservation model.
Further, preparation traffic signals described in step a identifies public data collection image, label and sets of video data
System environments is Windows10+Anaconda3+Tensorflow 1.5.0, and the original image vector number of generation is 6228,
Corresponding traffic lights location tags number is 6228, and test picture vector number is 2238, and corresponding traffic lights label number is 2940.
Further, change image size described in step b exports respectively for 640*480*3 is transformed to 128*128*3
The normalization numerical value in tri- channels each pixel RGB.
Further, coordinate label described in step c is upper left corner transverse and longitudinal coordinate and the lower right corner of traffic lights position rectangle
Transverse and longitudinal coordinate, it is described convert simultaneously normalized, as abscissa numerical value * according to the transformation rule of figure to coordinate label
0.2, Y value * 0.267, and transverse and longitudinal coordinate numerical value * 0.01 is normalized.
Further, according to the method for extracting the picture containing traffic lights information described in step d according to tab indexes
Label data centralized indexes extract the picture containing traffic lights, and remove the label of repeating label, leave behind an effective label
It is corresponded with picture, and the label of repeating part is summarized as test label data.
Further, the method for CNN network training described in step e and preservation model, comprising the following steps:
Step e1, input picture matrix shape is transformed to 128*128*3, data type float32 by 1*49152,
It is 1*4, data type float32 by input label matrixing;
Step e2, neural network model is established, network structure is five convolutional layers, and five pond layers are activated using ReLU
Function extracts image characteristic matrix, is sent into 60 groups of random data every time and is trained;
Step e3, using the method connected entirely by the image characteristic matrix extracted in step e2 using by the way of connecting entirely
Obtain final image feature vector P.
The present invention has the advantages that compared with the existing technology
The traffic light recognition method of the present invention provides a kind of multi-layer C NN network based on TensorFlow, the present invention adopt
Public data collection is identified with traffic signals, i.e. Traffic Lights Recognition (TLR) public benchmarks is most
For data set, have one containing the effective picture of 6228 width and corresponding a pair of of multi-tag, the characteristics of image for extracting the present invention
Fixed expressive faculty is labeled by convolutional neural networks algorithm to the traffic lights position in picture, is made of the invention effective
The accuracy for solving mark is not high and the slow technical problem of recognition speed.
Method proposed by the present invention has used multilayer convolutional neural networks to input figure compared to traditional images processing method
Piece carries out feature extraction, also brings better capability of fitting to model while obtaining image information more abundant, subtracts
Few over-fitting, keeps the development of computer vision field further.
Detailed description of the invention
Fig. 1 is multi-layer image convolution schematic diagram;
Fig. 2 is network structure flow chart;
Fig. 3 is recognition effect display diagram.
Specific embodiment
Below with reference to attached drawing, the present invention is described in detail.
Specific embodiment one
A kind of traffic light recognition method of the multi-layer C NN network based on TensorFlow, comprising the following steps:
Step a, prepare Traffic Lights Recognition (TLR) public benchmarks image, label and
Sets of video data;
Step b, to the image modification size of traffic signals identification public data concentration and its RGB tri- is exported with OpenCV
The normalization numerical value in channel;
Step c, convert simultaneously normalized according to the transformation rule of figure to coordinate label;
Step d, the picture containing traffic lights information is extracted according to tab indexes and correspond it with label;
Step e, picture and label are sent into CNN network and are trained simultaneously preservation model;
Specifically, the traffic light recognition method of a kind of multi-layer C NN network based on TensorFlow, described in step a
Data set prepares, and system experimentation environment is Windows10+Anaconda3+Tensorflow 1.5.0, the original image of generation
Vector number is 6228, and corresponding traffic lights location tags number is 6228, and test picture vector number is 2238, corresponding traffic lights
Label number is 2940.
Specifically, the traffic light recognition method of a kind of multi-layer C NN network based on TensorFlow, described in step b
Change image size and exports the normalization in each tri- channels pixel RGB respectively for 640*480*3 is transformed to 128*128*3
Numerical value.
Specifically, the traffic light recognition method of a kind of multi-layer C NN network based on TensorFlow, the corresponding mark of picture
Sign the upper left corner transverse and longitudinal coordinate and lower right corner transverse and longitudinal coordinate that information is traffic lights position rectangle, tag processes described in step c, for cross
Coordinate values * 0.2, Y value * 0.267, and transverse and longitudinal coordinate numerical value * 0.01 is normalized.
Specifically, the traffic light recognition method of a kind of multi-layer C NN network based on TensorFlow, described in step d,
According to label data centralized indexes, the picture containing traffic lights is extracted, and removes the label of repeating label, leaves behind one effectively
Label and picture correspond, and the label of repeating part is summarized as test label data.
Specifically, the traffic light recognition method of a kind of multi-layer C NN network based on TensorFlow, described in step e,
Simultaneously preservation model includes following steps to training CNN network:
Step e1, input picture matrix shape is transformed to 128*128*3, data type float32 by 1*49152,
It is 1*4, data type float32 by input label matrixing;
Step e2, neural network model is established, network structure is five convolutional layers, and five pond layers are activated using ReLU
Function extracts image characteristic matrix, is sent into 60 groups of random data every time and is trained;
Step e3, it to there is over-fitting when preventing and training, will be extracted in step e2 using the method connected entirely
Image characteristic matrix obtains final image feature vector P by the way of connecting entirely.
As shown in Figure 1, the present invention uses image convolution, i.e., convolution operation is done to digital picture, is in fact exactly to utilize convolution
Core (convolution mask) slides on the image, the grey scale pixel value in picture point is multiplied with the numerical value on corresponding convolution kernel, so
The value after all multiplications is added the gray value as pixel on the corresponding image of convolution kernel intermediate pixel, multilayer convolutional network afterwards
It can be realized more excellent nicety of grading.
As shown in Fig. 2, the present invention uses five layers of convolution, five layers of pond, the network model of two full articulamentums is finally directed to
Each group of training data exports the vector of a 1*4, indicates target traffic lights position.
Model measurement effect picture, as shown in Figure 3.
Claims (6)
1. a kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network, which is characterized in that including following step
It is rapid:
Step a, prepare traffic signals identification public data collection image, label and sets of video data;
Step b, to the image modification size of traffic signals identification public data concentration and tri- channels its RGB are exported with OpenCV
Normalization numerical value;
Step c, convert simultaneously normalized according to the transformation rule of figure to coordinate label;
Step d, the picture containing traffic lights information is extracted according to tab indexes and correspond it with label;
Step e, picture and label are sent into CNN network and are trained simultaneously preservation model.
2. a kind of traffic light recognition method based on TensorFlow combination multi-layer C NN network according to claim 1, special
Sign is that the system environments of the identification of preparation traffic signals described in step a public data collection image, label and sets of video data is
Windows10+Anaconda3+Tensorflow 1.5.0, the original image vector number of generation are 6228, corresponding traffic lights
Location tags number is 6228, and test picture vector number is 2238, and corresponding traffic lights label number is 2940.
3. a kind of traffic light recognition method of the multi-layer C NN network based on TensorFlow according to claim 2, feature
It is, change image size described in step b exports each pixel RGB tri- for 640*480*3 is transformed to 128*128*3 respectively
The normalization numerical value in a channel.
4. a kind of traffic light recognition method of the multi-layer C NN network based on TensorFlow according to claim 3, feature
It is, coordinate label described in step c is the upper left corner transverse and longitudinal coordinate and lower right corner transverse and longitudinal coordinate of traffic lights position rectangle, described
Convert simultaneously normalized, as abscissa numerical value * 0.2, ordinate number according to the transformation rule of figure to coordinate label
Value * 0.267, and transverse and longitudinal coordinate numerical value * 0.01 is normalized.
5. a kind of traffic light recognition method of the multi-layer C NN network based on TensorFlow according to claim 1, feature
It is, the method for the picture containing traffic lights information is extracted described in step d according to tab indexes to concentrate according to label data
Index extracts the picture containing traffic lights, and removes the label of repeating label, leaves behind an effective label and picture one is a pair of
It answers, and the label of repeating part is summarized as test label data.
6. a kind of traffic light recognition method of the multi-layer C NN network based on TensorFlow according to claim 1, feature
It is, the method for CNN network training described in step e and preservation model, comprising the following steps:
Step e1, input picture matrix shape is transformed to 128*128*3, data type float32 by 1*49152, it will be defeated
Enter label matrix and is transformed to 1*4, data type float32;
Step e2, neural network model is established, network structure is five convolutional layers, five pond layers, using ReLU activation primitive,
Image characteristic matrix is extracted, 60 groups of random data is sent into every time and is trained;
Step e3, the image characteristic matrix extracted in step e2 is used using the method connected entirely and is obtained by the way of connecting entirely
Final image feature vector P.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811167453.8A CN109508635B (en) | 2018-10-08 | 2018-10-08 | Traffic light identification method based on TensorFlow combined with multilayer CNN network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811167453.8A CN109508635B (en) | 2018-10-08 | 2018-10-08 | Traffic light identification method based on TensorFlow combined with multilayer CNN network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109508635A true CN109508635A (en) | 2019-03-22 |
CN109508635B CN109508635B (en) | 2022-01-07 |
Family
ID=65746279
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811167453.8A Expired - Fee Related CN109508635B (en) | 2018-10-08 | 2018-10-08 | Traffic light identification method based on TensorFlow combined with multilayer CNN network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109508635B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104050827A (en) * | 2014-06-06 | 2014-09-17 | 北京航空航天大学 | Traffic signal lamp automatic detection and recognition method based on visual sense |
US20170083792A1 (en) * | 2015-09-22 | 2017-03-23 | Xerox Corporation | Similarity-based detection of prominent objects using deep cnn pooling layers as features |
CN106650641A (en) * | 2016-12-05 | 2017-05-10 | 北京文安智能技术股份有限公司 | Traffic light positioning and identification method, device and system |
CN106845335A (en) * | 2016-11-29 | 2017-06-13 | 歌尔科技有限公司 | Gesture identification method, device and virtual reality device for virtual reality device |
CN107220643A (en) * | 2017-04-12 | 2017-09-29 | 广东工业大学 | The Traffic Sign Recognition System of deep learning model based on neurological network |
CN108052933A (en) * | 2018-01-16 | 2018-05-18 | 杭州国辰机器人科技有限公司 | Road Identification system and method based on convolutional neural networks |
CN108090411A (en) * | 2016-11-23 | 2018-05-29 | 福特全球技术公司 | Traffic lights detection and classification are carried out using computer vision and deep learning |
CN108335510A (en) * | 2018-03-21 | 2018-07-27 | 北京百度网讯科技有限公司 | Traffic lights recognition methods, device and equipment |
-
2018
- 2018-10-08 CN CN201811167453.8A patent/CN109508635B/en not_active Expired - Fee Related
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104050827A (en) * | 2014-06-06 | 2014-09-17 | 北京航空航天大学 | Traffic signal lamp automatic detection and recognition method based on visual sense |
US20170083792A1 (en) * | 2015-09-22 | 2017-03-23 | Xerox Corporation | Similarity-based detection of prominent objects using deep cnn pooling layers as features |
CN108090411A (en) * | 2016-11-23 | 2018-05-29 | 福特全球技术公司 | Traffic lights detection and classification are carried out using computer vision and deep learning |
CN106845335A (en) * | 2016-11-29 | 2017-06-13 | 歌尔科技有限公司 | Gesture identification method, device and virtual reality device for virtual reality device |
CN106650641A (en) * | 2016-12-05 | 2017-05-10 | 北京文安智能技术股份有限公司 | Traffic light positioning and identification method, device and system |
CN107220643A (en) * | 2017-04-12 | 2017-09-29 | 广东工业大学 | The Traffic Sign Recognition System of deep learning model based on neurological network |
CN108052933A (en) * | 2018-01-16 | 2018-05-18 | 杭州国辰机器人科技有限公司 | Road Identification system and method based on convolutional neural networks |
CN108335510A (en) * | 2018-03-21 | 2018-07-27 | 北京百度网讯科技有限公司 | Traffic lights recognition methods, device and equipment |
Also Published As
Publication number | Publication date |
---|---|
CN109508635B (en) | 2022-01-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021022970A1 (en) | Multi-layer random forest-based part recognition method and system | |
CN106599773B (en) | Deep learning image identification method and system for intelligent driving and terminal equipment | |
CN109859158A (en) | A kind of detection system, method and the machinery equipment on the working region boundary of view-based access control model | |
CN106845487A (en) | A kind of licence plate recognition method end to end | |
CN106096602A (en) | Chinese license plate recognition method based on convolutional neural network | |
CN109509156B (en) | Image defogging processing method based on generation countermeasure model | |
CN111008632B (en) | License plate character segmentation method based on deep learning | |
CN109410144B (en) | End-to-end image defogging processing method based on deep learning | |
CN111062381B (en) | License plate position detection method based on deep learning | |
CN110827312B (en) | Learning method based on cooperative visual attention neural network | |
WO2021142944A1 (en) | Vehicle behaviour recognition method and apparatus | |
CN112712052A (en) | Method for detecting and identifying weak target in airport panoramic video | |
CN109657538B (en) | Scene segmentation method and system based on context information guidance | |
CN112487981A (en) | MA-YOLO dynamic gesture rapid recognition method based on two-way segmentation | |
CN109325407B (en) | Optical remote sensing video target detection method based on F-SSD network filtering | |
CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
CN112464731A (en) | Traffic sign detection and identification method based on image processing | |
GUO et al. | Traffic signs recognition based on visual attention mechanism | |
CN109753962A (en) | Text filed processing method in natural scene image based on hybrid network | |
CN113077438B (en) | Cell nucleus region extraction method and imaging method for multi-cell nucleus color image | |
CN114445620A (en) | Target segmentation method for improving Mask R-CNN | |
CN113902753A (en) | Image semantic segmentation method and system based on dual-channel and self-attention mechanism | |
CN109034007A (en) | A kind of vegetation extraction optimization method of the remote sensing image based on convolutional neural networks | |
CN111881914B (en) | License plate character segmentation method and system based on self-learning threshold | |
CN110334703B (en) | Ship detection and identification method in day and night image |
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 | ||
TA01 | Transfer of patent application right |
Effective date of registration: 20211221 Address after: 570000 99 Longkun South Road, Qiongshan District, Haikou City, Hainan Province Applicant after: HAINAN NORMAL University Address before: 150080 No. 52, Xuefu Road, Nangang District, Heilongjiang, Harbin Applicant before: HARBIN University OF SCIENCE AND TECHNOLOGY |
|
TA01 | Transfer of patent application right | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220107 |
|
CF01 | Termination of patent right due to non-payment of annual fee |