CN106909886B - A kind of high-precision method for traffic sign detection and system based on deep learning - Google Patents
A kind of high-precision method for traffic sign detection and system based on deep learning Download PDFInfo
- Publication number
- CN106909886B CN106909886B CN201710041906.1A CN201710041906A CN106909886B CN 106909886 B CN106909886 B CN 106909886B CN 201710041906 A CN201710041906 A CN 201710041906A CN 106909886 B CN106909886 B CN 106909886B
- Authority
- CN
- China
- Prior art keywords
- traffic sign
- feature
- training
- images
- convolutional neural
- 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.)
- Active
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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of high-precision method for traffic sign detection and system based on deep learning.Wherein, this method combines depth learning technology with high-precision road traffic sign detection technology, by being trained to SSD network and convolutional neural networks, the traffic sign feature after carrying out overlapping cutting in proportion in video flowing is extracted using trained SSD network, the traffic sign feature extracted according to SSD network, the feature of traffic sign feature is extracted using trained convolutional neural networks, the feature for the traffic sign feature extracted is matched with the feature of the positive and negative two classes traffic sign of Traffic Sign Images Test database, retain positive class traffic sign feature, obtain high-precision traffic sign matching the selection result, effectively improve the accuracy rate of high-precision road traffic sign detection.
Description
Technical field
The invention belongs to artificial intelligence field more particularly to a kind of high-precision road traffic sign detection sides based on deep learning
Method and system.
Background technique
Deep learning is the topnotch of current machine learning development, a kind of side of the convolutional neural networks as deep learning
Method has preferable effect in fields such as object identification, image procossings.For feature extraction, convolutional neural networks have permissible
The advantage of automatic study characteristics of image, reduces manual intervention, extracts the feature of high quality, thus to improve images match
Accuracy rate lays a solid foundation.
Since the method for deep learning does not do enough specific aim treatment of details in image pre-processing module, image is excessive
The effect of expected high-precision testing result may be not achieved in target object proportion when too small, in order to improve detection identification
Accuracy, it usually needs specific application scenarios and picture material are analyzed and carry out image preprocessing etc..
Traffic sign is as one of the important goal in video monitoring, and to subsequent identification, auxiliary positioning is led for accurate detection
Boat plays conclusive effect.The huge number of traffic sign, size, angle are disobeyed, itself are difficult to accomplish accurately to detect, and
And in true environment, influenced by factors such as weather, illumination, so that the detection of traffic sign is more difficult, especially
In automatic Pilot scene, the detection and identification of traffic sign play vital work to the understanding of driving ambient enviroment for it
With.Such as the speed etc. of current vehicle is controlled by detection identification speed(-)limit sign；On the other hand, traffic sign is embedded into height
In precision map, crucial booster action is also functioned to location navigation.
In view of this, being badly in need of solving the problems, such as lower to the accuracy rate of road traffic sign detection.
Summary of the invention
Lower to the accuracy rate of road traffic sign detection in order to solve the problems, such as, the first object of the present invention is to provide a kind of base
In the high-precision method for traffic sign detection of deep learning.This method greatly increases the accuracy rate and detection essence of feature extraction
Degree.
A kind of high-precision method for traffic sign detection based on deep learning of the invention, comprising:
Step 1: it acquires the Traffic Sign Images of history and carries out overlapping cutting in proportion, then be input to SSD network, until
Obtain optimal SSD network parameter；
Step 2: extracted using the optimal SSD network of parameter cutting after Traffic Sign Images traffic sign feature and by
According to whether being that traffic sign feature is divided into these two types of traffic sign features of positive and negative classification and stores to Traffic Sign Images testing number
According to library；
Step 3: using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolution
Neural network parameter；
Step 4: the traffic indication map after overlapping is cut in proportion in video flowing is extracted using the optimal SSD network of parameter
The traffic sign feature of picture, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract
The feature of traffic sign feature out；
Step 5: by the feature for the traffic sign feature extracted and Traffic Sign Images Test database it is positive and negative this two
Class traffic sign feature is matched respectively, and retains positive class traffic sign feature；
Step 6: in the Traffic Sign Images before final testing result to be restored to overlapping cutting in proportion.
Further, in the step 1, the Traffic Sign Images of collected history are subjected to overlapping cutting in proportion
Later, position coordinates of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after saving cutting.
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, and save the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in Traffic Sign Images are that accurate reproduction Traffic Sign Images improve coordinate basis.
Further, it during the step 1 training SSD network, is trained first using default parameters, according to
Training intermediate result is constantly adjusted initial weight, training rate and the number of iterations, until the SSD network is with default
Efficiency reach preset recognition effect.
SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the friendship after overlapping cutting in proportion
The traffic sign feature of logical sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
Further, it during step 3 training convolutional neural networks, is instructed first using default parameters
Practice, according to training intermediate result, initial weight, training rate and the number of iterations are constantly adjusted, until the convolution mind
Preset recognition effect is reached with preset efficiency through network.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
The second object of the present invention is to provide a kind of high-precision road traffic sign detection system based on deep learning.
A kind of high-precision road traffic sign detection system based on deep learning of the invention, comprising:
SSD network training module is used to acquire the Traffic Sign Images of history and carries out overlapping cutting in proportion, then defeated
Enter to SSD network, until obtaining optimal SSD network parameter；
Tagsort module is used to extract the friendship of Traffic Sign Images after cutting using the optimal SSD network of parameter
Logical flag sign and according to whether being divided into positive and negative classification these two types traffic sign feature for traffic sign feature and store to traffic
Sign image Test database；
Convolutional neural networks training module is used for using Traffic Sign Images Test database come training convolutional nerve net
Network, until obtaining optimal convolutional neural networks parameter；
Characteristic extracting module is used to be extracted using the optimal SSD network of parameter in video flowing and is overlapped cutting in proportion
The traffic sign feature of Traffic Sign Images afterwards, then the traffic sign feature of extraction is input to the optimal convolutional Neural of parameter
Network, and then extract the feature of traffic sign feature；
Characteristic matching module, the feature and Traffic Sign Images detection data of the traffic sign feature for being used to extract
Positive and negative these two types traffic sign feature is matched respectively in library, and retains positive class traffic sign feature；
Image restoring module, the traffic indication map before being used to for final testing result being restored to overlapping cutting in proportion
As in.
Further, which further includes cutting traffic indication map image position logging modle, is used to go through collected
After the Traffic Sign Images of history carry out overlapping cutting in proportion, the Traffic Sign Images after saving cutting are before overlapping cutting
Position coordinates in Traffic Sign Images.
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, and save the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in Traffic Sign Images are that accurate reproduction Traffic Sign Images improve coordinate basis.
Further, it in the SSD network training module, is trained first using default parameters, according in training
Between as a result, to initial weight, training rate and the number of iterations be constantly adjusted, until the SSD network is with preset efficiency
Reach preset recognition effect.
SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the friendship after overlapping cutting in proportion
The traffic sign feature of logical sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
Further, it in the convolutional neural networks training module, is trained first using default parameters, according to instruction
Practice intermediate result, initial weight, training rate and the number of iterations is constantly adjusted, until the convolutional neural networks are with pre-
If efficiency reach preset recognition effect.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
The second object of the present invention is to provide a kind of high-precision road traffic sign detection system based on deep learning.
High-precision road traffic sign detection system the present invention also provides another kind based on deep learning.
Another high-precision road traffic sign detection system based on deep learning of the invention, comprising:
Image collecting device, is configured as the Traffic Sign Images of acquisition history, and is sent to server；
The server, is configured as:
Overlapping cutting is carried out by the Traffic Sign Images of the history of acquisition and in proportion, then is input to SSD network, until
To optimal SSD network parameter；
Extracted using the optimal SSD network of parameter cutting after Traffic Sign Images traffic sign feature and according to whether
It is divided into these two types of traffic sign features of positive and negative classification for traffic sign feature and stores to Traffic Sign Images Test database；
Using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolutional Neural net
Network parameter；
The friendship of the Traffic Sign Images after overlapping is cut in proportion in video flowing is extracted using the optimal SSD network of parameter
Logical flag sign, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract traffic
The feature of flag sign；
By these two types of traffic positive and negative in the feature for the traffic sign feature extracted and Traffic Sign Images Test database
Flag sign is matched respectively, and retains positive class traffic sign feature；
In Traffic Sign Images before final testing result to be restored to overlapping cutting in proportion.
Further, the server is also configured in proportion carry out the Traffic Sign Images of collected history
Position coordinates of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after overlapping cutting, after saving cutting.
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, and save the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in Traffic Sign Images are that accurate reproduction Traffic Sign Images improve coordinate basis.
Further, the server is also configured to use default parameters first during training SSD network
It is trained, according to training intermediate result, initial weight, training rate and the number of iterations is constantly adjusted, until described
SSD network reaches preset recognition effect with preset efficiency.
SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the friendship after overlapping cutting in proportion
The traffic sign feature of logical sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
Further, the server, during being also configured to training convolutional neural networks, first using default
Parameter is trained, and according to training intermediate result, is constantly adjusted to initial weight, training rate and the number of iterations, until
The convolutional neural networks reach preset recognition effect with preset efficiency.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
SSD network of the present invention, full name in English are as follows: Single Shot MultiBox Detector network,
It is after carrying out convolution to image using single convolutional neural networks, prediction is a series of not at each position of characteristic image
With the bounding box of size and length-width ratio.In test phase, SSD network is to the object for separately including each classification in each bounding box
A possibility that body, is predicted, and is adjusted to bounding box to adapt to the shape of target object.
Compared with prior art, the beneficial effects of the present invention are:
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, by SSD network and convolutional neural networks to Traffic Sign Images data
Feature extraction is carried out, and the testing result by obtaining after SSD training refreshes high-precision Traffic Sign Images database, greatly
Improve the accuracy rate and detection accuracy of feature extraction.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is a kind of process signal of the high-precision method for traffic sign detection based on deep learning in the embodiment of the present invention
Figure；
Fig. 2 is a kind of high-precision road traffic sign detection system structure signal based on deep learning in the embodiment of the present invention
Figure；
Fig. 3 is that another kind is illustrated based on the high-precision road traffic sign detection system structure of deep learning in the embodiment of the present invention
Figure.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
As background technique is introduced, the problem lower to the accuracy rate of road traffic sign detection exists in the prior art,
In order to solve technical problem as above, the invention proposes a kind of high-precision method for traffic sign detection based on deep learning.
Fig. 1 is a kind of process signal of the high-precision method for traffic sign detection based on deep learning in the embodiment of the present invention
Figure, high-precision method for traffic sign detection of one of the present embodiment as shown in the figure based on deep learning may include:
S101 acquires the Traffic Sign Images of history and carries out overlapping cutting in proportion, then is input to SSD network, until
Obtain optimal SSD network parameter.
In the specific implementation, acquiring the Traffic Sign Images of history and carrying out overlapping cutting in proportion, after cutting being overlapped
Traffic Sign Images are stored to historical traffic sign image database.Then it is selected out of historical traffic sign image database again
Take the sample of trained SSD network, the sample set of composing training SSD network.
It during training SSD network, is trained first using default parameters, according to training intermediate result, to first
Beginning weight, training rate and the number of iterations are constantly adjusted, until the SSD network reaches preset knowledge with preset efficiency
Other effect.SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the traffic after overlapping cutting in proportion
The traffic sign feature of sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
S102, extracted using the optimal SSD network of parameter cutting after Traffic Sign Images traffic sign feature and by
According to whether being that traffic sign feature is divided into these two types of traffic sign features of positive and negative classification and stores to Traffic Sign Images testing number
According to library.
In the specific implementation, the traffic sign for extracting Traffic Sign Images after cutting using the optimal SSD network of parameter is special
It levies and stores to Traffic Sign Images Test database；For the traffic sign feature extracted using the optimal SSD network of parameter,
If traffic sign feature, then the class that is positive traffic sign feature；If not traffic sign feature, then the class that is negative traffic sign is special
Sign.
Wherein, these two types of traffic sign features of positive and negative classification are stored to Traffic Sign Images Test database.
S103, using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolution
Neural network parameter.
In the specific implementation, the training set of training convolutional neural networks is chosen from Traffic Sign Images Test database, and
It is input in convolutional neural networks.Wherein, convolutional neural networks from be input to output include convolutional layer, pond layer, articulamentum and
Softmax returns classifier layer.Softmax returns classifier layer and there was only one layer, and the number of plies of convolutional layer, pond layer and articulamentum
It is set according to actual conditions or adjusts.Such as: convolutional neural networks include three-layer coil lamination, three layers of pond layer, three layers of full connection
Layer and classifier layer is returned positioned at last softmax.
During hands-on convolutional neural networks, it is trained first using default parameters, it is intermediate according to training
As a result, being constantly adjusted to initial weight, training rate and the number of iterations, until the convolutional neural networks are with preset effect
Rate reaches preset recognition effect.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
S104 extracts the traffic indication map after overlapping is cut in proportion in video flowing using the optimal SSD network of parameter
The traffic sign feature of picture, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract
The feature of traffic sign feature out.
In the specific implementation, after carrying out overlapping cutting in proportion in video flowing using the extraction of trained SSD network
Traffic sign feature, the traffic sign feature extracted according to SSD network, using trained convolutional neural networks to extraction band
Feature carry out further training screening.
S105, by positive and negative these two types in the feature for the traffic sign feature extracted and Traffic Sign Images Test database
Traffic sign feature is matched respectively, and retains positive class traffic sign feature.
In the specific implementation, if the feature for the traffic sign feature extracted and positive class traffic sign characteristic matching, are extracted
To the feature of traffic sign feature retain the traffic sign feature, and then detect Traffic Sign Images.
Final testing result is restored in the Traffic Sign Images before overlapping is cut by S106 in proportion.
In another embodiment, it after the Traffic Sign Images of collected history being carried out overlapping cutting in proportion, protects
Position coordinates of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after depositing cutting.
In this way by using there is the cutting method of overlapping to cut image, the ratio that target object accounts for original image is increased
Example can effectively improve the accuracy rate of feature extraction, and save friendship of the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in logical sign image are that accurate reproduction Traffic Sign Images improve coordinate basis.
The present embodiment combines depth learning technology with high-precision road traffic sign detection technology, by SSD network and
Convolutional neural networks are trained, and are extracted after carrying out overlapping cutting in proportion in video flowing using trained SSD network
Traffic sign feature, the traffic sign feature extracted according to SSD network, utilize trained convolutional neural networks extract hand over
The feature of logical flag sign, by the feature for the traffic sign feature extracted and positive and negative the two of Traffic Sign Images Test database
The feature of class traffic sign is matched, and positive class traffic sign feature is retained, and obtains high-precision traffic sign matching the selection result,
Effectively improve the accuracy rate of high-precision road traffic sign detection.
Fig. 2 is a kind of high-precision road traffic sign detection system structure signal based on deep learning in the embodiment of the present invention
Figure, high-precision road traffic sign detection system of one of the present embodiment as shown in the figure based on deep learning may include:
(1) SSD network training module is used to acquire the Traffic Sign Images of history and carries out overlapping cutting in proportion,
It is input to SSD network again, until obtaining optimal SSD network parameter.
In the specific implementation, acquiring the Traffic Sign Images of history and carrying out overlapping cutting in proportion, after cutting being overlapped
Traffic Sign Images are stored to historical traffic sign image database.Then it is selected out of historical traffic sign image database again
Take the sample of trained SSD network, the sample set of composing training SSD network.
It during training SSD network, is trained first using default parameters, according to training intermediate result, to first
Beginning weight, training rate and the number of iterations are constantly adjusted, until the SSD network reaches preset knowledge with preset efficiency
Other effect.SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the traffic after overlapping cutting in proportion
The traffic sign feature of sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
(2) tagsort module is used to extract Traffic Sign Images after cutting using the optimal SSD network of parameter
Traffic sign feature and according to whether being divided into these two types of traffic sign features of positive and negative classification for traffic sign feature and store best friend
Logical sign image Test database.
In the specific implementation, the traffic sign for extracting Traffic Sign Images after cutting using the optimal SSD network of parameter is special
It levies and stores to Traffic Sign Images Test database；For the traffic sign feature extracted using the optimal SSD network of parameter,
If traffic sign feature, then the class that is positive traffic sign feature；If not traffic sign feature, then the class that is negative traffic sign is special
Sign.
Wherein, these two types of traffic sign features of positive and negative classification are stored to Traffic Sign Images Test database.
(3) convolutional neural networks training module is used for using Traffic Sign Images Test database come training convolutional mind
Through network, until obtaining optimal convolutional neural networks parameter.
In the specific implementation, the training set of training convolutional neural networks is chosen from Traffic Sign Images Test database, and
It is input in convolutional neural networks.Wherein, convolutional neural networks from be input to output include convolutional layer, pond layer, articulamentum and
Softmax returns classifier layer.Softmax returns classifier layer and there was only one layer, and the number of plies of convolutional layer, pond layer and articulamentum
It is set according to actual conditions or adjusts.Such as: convolutional neural networks include three-layer coil lamination, three layers of pond layer, three layers of full connection
Layer and classifier layer is returned positioned at last softmax.
During hands-on convolutional neural networks, it is trained first using default parameters, it is intermediate according to training
As a result, being constantly adjusted to initial weight, training rate and the number of iterations, until the convolutional neural networks are with preset effect
Rate reaches preset recognition effect.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
(4) characteristic extracting module is used to be extracted using the optimal SSD network of parameter in video flowing to be overlapped in proportion and be cut
The traffic sign feature of Traffic Sign Images after cutting, then the traffic sign feature of extraction is input to the optimal convolution mind of parameter
Through network, and then extract the feature of traffic sign feature.
In the specific implementation, after carrying out overlapping cutting in proportion in video flowing using the extraction of trained SSD network
Traffic sign feature, the traffic sign feature extracted according to SSD network, using trained convolutional neural networks to extraction band
Feature carry out further training screening.
(5) characteristic matching module, the feature and Traffic Sign Images for the traffic sign feature for being used to extract detect
Positive and negative these two types traffic sign feature is matched respectively in database, and retains positive class traffic sign feature.
In the specific implementation, if the feature for the traffic sign feature extracted and positive class traffic sign characteristic matching, are extracted
To the feature of traffic sign feature retain the traffic sign feature, and then detect Traffic Sign Images.
(6) image restoring module, the traffic mark before being used to for final testing result being restored to overlapping cutting in proportion
In will image.
In another embodiment, which further includes cutting traffic indication map image position logging modle, is used to acquire
To history Traffic Sign Images carry out overlapping cutting in proportion after, save cutting after Traffic Sign Images cut in overlapping
The position coordinates in Traffic Sign Images before cutting.
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, and save the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in Traffic Sign Images are that accurate reproduction Traffic Sign Images improve coordinate basis.
The present embodiment combines depth learning technology with high-precision road traffic sign detection technology, by SSD network and
Convolutional neural networks are trained, and are extracted after carrying out overlapping cutting in proportion in video flowing using trained SSD network
Traffic sign feature, the traffic sign feature extracted according to SSD network, utilize trained convolutional neural networks extract hand over
The feature of logical flag sign, by the feature for the traffic sign feature extracted and positive and negative the two of Traffic Sign Images Test database
The feature of class traffic sign is matched, and positive class traffic sign feature is retained, and obtains high-precision traffic sign matching the selection result,
Effectively improve the accuracy rate of high-precision road traffic sign detection.
Fig. 3 is that another kind is illustrated based on the high-precision road traffic sign detection system structure of deep learning in the embodiment of the present invention
Figure, high-precision road traffic sign detection system of one of the present embodiment as shown in the figure based on deep learning may include:
(1) image collecting device, is configured as the Traffic Sign Images of acquisition history, and is sent to server.
Wherein, image collecting device can be realized using video camera, be used to acquire traffic mark image.
(2) server is configured as:
Overlapping cutting is carried out by the Traffic Sign Images of the history of acquisition and in proportion, then is input to SSD network, until
To optimal SSD network parameter；
Extracted using the optimal SSD network of parameter cutting after Traffic Sign Images traffic sign feature and according to whether
It is divided into these two types of traffic sign features of positive and negative classification for traffic sign feature and stores to Traffic Sign Images Test database；
Using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolutional Neural net
Network parameter；
The friendship of the Traffic Sign Images after overlapping is cut in proportion in video flowing is extracted using the optimal SSD network of parameter
Logical flag sign, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract traffic
The feature of flag sign；
By these two types of traffic positive and negative in the feature for the traffic sign feature extracted and Traffic Sign Images Test database
Flag sign is matched respectively, and retains positive class traffic sign feature；
In Traffic Sign Images before final testing result to be restored to overlapping cutting in proportion.
In another embodiment, the server, be also configured to by the Traffic Sign Images of collected history by than
Position of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after example carries out overlapping cutting, after saving cutting
Coordinate.
The present invention increases target object and accounts for original image by using there is the cutting method of overlapping to cut image
Ratio can effectively improve the accuracy rate of feature extraction, and save the Traffic Sign Images after cutting before overlapping cutting
Position coordinates in Traffic Sign Images are that accurate reproduction Traffic Sign Images improve coordinate basis.
In another embodiment, the server is also configured to during training SSD network, first using silent
Recognize parameter to be trained, according to training intermediate result, initial weight, training rate and the number of iterations are constantly adjusted, directly
Preset recognition effect is reached with preset efficiency to the SSD network.
SSD network parameter that in this way can be optimal, and then accurately extract in video flowing the friendship after overlapping cutting in proportion
The traffic sign feature of logical sign image is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
In another embodiment, the server during being also configured to training convolutional neural networks, makes first
It is trained with default parameters, according to training intermediate result, initial weight, training rate and the number of iterations is constantly adjusted
It is whole, until the convolutional neural networks reach preset recognition effect with preset efficiency.
Convolutional neural networks parameter that in this way can be optimal, and then the traffic sign for accurately extracting Traffic Sign Images is special
The feature of sign is the accuracy rate of traffic sign feature extraction and the basis that detection accuracy is established.
The present embodiment combines depth learning technology with high-precision road traffic sign detection technology, by SSD network and
Convolutional neural networks are trained, and are extracted after carrying out overlapping cutting in proportion in video flowing using trained SSD network
Traffic sign feature, the traffic sign feature extracted according to SSD network, utilize trained convolutional neural networks extract hand over
The feature of logical flag sign, by the feature for the traffic sign feature extracted and positive and negative the two of Traffic Sign Images Test database
The feature of class traffic sign is matched, and positive class traffic sign feature is retained, and obtains high-precision traffic sign matching the selection result,
Effectively improve the accuracy rate of high-precision road traffic sign detection.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the present invention
Formula.Moreover, the present invention, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random
AccessMemory, RAM) etc..
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (10)
1. a kind of high-precision method for traffic sign detection based on deep learning characterized by comprising
Step 1: acquiring the Traffic Sign Images of history and carry out overlapping cutting in proportion, then be input to SSD network, until obtaining
Optimal SSD network parameter；
Step 2: the traffic sign feature of Traffic Sign Images is extracted after cutting using the optimal SSD network of parameter and according to being
It is no to be divided into these two types of traffic sign features of positive and negative classification for traffic sign feature and store to Traffic Sign Images Test database；
Step 3: using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolutional Neural
Network parameter；
Step 4: the Traffic Sign Images after overlapping is cut in proportion in video flowing are extracted using the optimal SSD network of parameter
Traffic sign feature, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract friendship
The feature of logical flag sign；
Step 5: by these two types of friendships positive and negative in the feature for the traffic sign feature extracted and Traffic Sign Images Test database
Logical flag sign is matched respectively, and retains positive class traffic sign feature；
Step 6: in the Traffic Sign Images before final testing result to be restored to overlapping cutting in proportion.
2. the high-precision method for traffic sign detection based on deep learning as described in claim 1, which is characterized in that described
Traffic mark in step 1, after the Traffic Sign Images of collected history are carried out overlapping cutting in proportion, after saving cutting
Position coordinates of the will image in the Traffic Sign Images before overlapping cutting.
3. the high-precision method for traffic sign detection based on deep learning as described in claim 1, which is characterized in that described
It during step 1 training SSD network, is trained first using default parameters, according to training intermediate result, to initial power
Value, training rate and the number of iterations are constantly adjusted, until the SSD network reaches preset identification effect with preset efficiency
Fruit.
4. the high-precision method for traffic sign detection based on deep learning as described in claim 1, which is characterized in that described
It during step 3 training convolutional neural networks, is trained first using default parameters, according to training intermediate result, to first
Beginning weight, training rate and the number of iterations are constantly adjusted, until the convolutional neural networks reach pre- with preset efficiency
If recognition effect.
5. a kind of high-precision road traffic sign detection system based on deep learning characterized by comprising
SSD network training module is used to acquire the Traffic Sign Images of history and carries out overlapping cutting in proportion, then is input to
SSD network, until obtaining optimal SSD network parameter；
Tagsort module is used to extract the traffic mark of Traffic Sign Images after cutting using the optimal SSD network of parameter
Will feature and according to whether being divided into these two types of traffic sign features of positive and negative classification for traffic sign feature and store to traffic sign
Image inspection data library；
Convolutional neural networks training module is used for using Traffic Sign Images Test database come training convolutional neural networks,
Until obtaining optimal convolutional neural networks parameter；
Characteristic extracting module is used to be extracted using the optimal SSD network of parameter in video flowing after being overlapped cutting in proportion
The traffic sign feature of Traffic Sign Images, then the traffic sign feature of extraction is input to the optimal convolutional Neural net of parameter
Network, and then extract the feature of traffic sign feature；
In characteristic matching module, the feature of the traffic sign feature for being used to extract and Traffic Sign Images Test database
Positive and negative these two types traffic sign feature is matched respectively, and retains positive class traffic sign feature；
Image restoring module, the Traffic Sign Images before being used to for final testing result being restored to overlapping cutting in proportion
In.
6. a kind of high-precision road traffic sign detection system based on deep learning as claimed in claim 5, which is characterized in that should
System further includes position logging modle, is used in proportion carry out the Traffic Sign Images of collected history overlapping cutting
Afterwards, position coordinates of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after saving cutting.
7. a kind of high-precision road traffic sign detection system based on deep learning as claimed in claim 5, which is characterized in that
It in the SSD network training module, is trained first using default parameters, according to training intermediate result, to initial weight, instruction
Practice rate and the number of iterations is constantly adjusted, until the SSD network reaches preset recognition effect with preset efficiency；
Or in the convolutional neural networks training module, be trained first using default parameters, according to training intermediate result,
Initial weight, training rate and the number of iterations are constantly adjusted, until the convolutional neural networks are reached with preset efficiency
To preset recognition effect.
8. a kind of high-precision road traffic sign detection system based on deep learning characterized by comprising
Image collecting device, is configured as the Traffic Sign Images of acquisition history, and is sent to server；
The server, is configured as:
Overlapping cutting is carried out by the Traffic Sign Images of the history of acquisition and in proportion, then is input to SSD network, until obtaining most
Excellent SSD network parameter；
The SSD network for using parameter optimal extract cutting after Traffic Sign Images traffic sign feature and according to whether for friendship
Logical flag sign is divided into these two types of traffic sign features of positive and negative classification and stores to Traffic Sign Images Test database；
Using Traffic Sign Images Test database come training convolutional neural networks, until obtaining optimal convolutional neural networks ginseng
Number；
The traffic mark of the Traffic Sign Images after overlapping is cut in proportion in video flowing is extracted using the optimal SSD network of parameter
Will feature, then the traffic sign feature of extraction is input to the optimal convolutional neural networks of parameter, and then extract traffic sign
The feature of feature；
By these two types of traffic signs positive and negative in the feature for the traffic sign feature extracted and Traffic Sign Images Test database
Feature is matched respectively, and retains positive class traffic sign feature；
In Traffic Sign Images before final testing result to be restored to overlapping cutting in proportion.
9. a kind of high-precision road traffic sign detection system based on deep learning as claimed in claim 8, which is characterized in that institute
Server is stated, is also configured to after the Traffic Sign Images of collected history are carried out overlapping cutting in proportion, preservation is cut
Position coordinates of the Traffic Sign Images in the Traffic Sign Images before overlapping cutting after cutting.
10. a kind of high-precision road traffic sign detection system based on deep learning as claimed in claim 8, which is characterized in that
The server is also configured to be trained using default parameters first, during training SSD network according to training
Intermediate result is constantly adjusted initial weight, training rate and the number of iterations, until the SSD network is with preset effect
Rate reaches preset recognition effect；
Or the server, during being also configured to training convolutional neural networks, instructed first using default parameters
Practice, according to training intermediate result, initial weight, training rate and the number of iterations are constantly adjusted, until the convolution mind
Preset recognition effect is reached with preset efficiency through network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041906.1A CN106909886B (en) | 2017-01-20 | 2017-01-20 | A kind of high-precision method for traffic sign detection and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710041906.1A CN106909886B (en) | 2017-01-20 | 2017-01-20 | A kind of high-precision method for traffic sign detection and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106909886A CN106909886A (en) | 2017-06-30 |
| CN106909886B true CN106909886B (en) | 2019-05-03 |
Family
ID=59207028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710041906.1A Active CN106909886B (en) | 2017-01-20 | 2017-01-20 | A kind of high-precision method for traffic sign detection and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106909886B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108230421A (en) * | 2017-09-19 | 2018-06-29 | 北京市商汤科技开发有限公司 | A kind of road drawing generating method, device, electronic equipment and computer storage media |
| CN107909021A (en) * | 2017-11-07 | 2018-04-13 | 浙江师范大学 | A kind of guideboard detection method based on single deep layer convolutional neural networks |
| CN108009518A (en) * | 2017-12-19 | 2018-05-08 | 大连理工大学 | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks |
| CN108615230A (en) * | 2018-03-16 | 2018-10-02 | 北京邮电大学 | A kind of hub surface method for detecting abnormality and system |
| JP6985207B2 (en) * | 2018-05-09 | 2021-12-22 | トヨタ自動車株式会社 | Autonomous driving system |
| CN109447066B (en) * | 2018-10-18 | 2021-08-20 | 中国人民武装警察部队海警学院 | Rapid and accurate single-stage target detection method and device |
| JP6921233B2 (en) * | 2019-01-11 | 2021-08-18 | アドバンスド ニュー テクノロジーズ カンパニー リミテッド | Logistic regression modeling method using secret sharing |
| CN112784084B (en) * | 2019-11-08 | 2024-01-26 | 阿里巴巴集团控股有限公司 | Image processing method and device and electronic equipment |
| CN111243310B (en) * | 2020-01-10 | 2021-08-31 | 广州大学 | Traffic sign recognition method, system, medium, and apparatus |
| US11887379B2 (en) | 2021-08-17 | 2024-01-30 | Robert Bosch Gmbh | Road sign content prediction and search in smart data management for training machine learning model |
| US11978264B2 (en) | 2021-08-17 | 2024-05-07 | Robert Bosch Gmbh | Scalable road sign knowledge graph construction with machine learning and human in the loop |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930830A (en) * | 2016-05-18 | 2016-09-07 | 大连理工大学 | Road surface traffic sign recognition method based on convolution neural network |
| CN106022300A (en) * | 2016-06-02 | 2016-10-12 | 中国科学院信息工程研究所 | Traffic sign identifying method and traffic sign identifying system based on cascading deep learning |
| CN106326858A (en) * | 2016-08-23 | 2017-01-11 | 北京航空航天大学 | Road traffic sign automatic identification and management system based on deep learning |
-
2017
- 2017-01-20 CN CN201710041906.1A patent/CN106909886B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105930830A (en) * | 2016-05-18 | 2016-09-07 | 大连理工大学 | Road surface traffic sign recognition method based on convolution neural network |
| CN106022300A (en) * | 2016-06-02 | 2016-10-12 | 中国科学院信息工程研究所 | Traffic sign identifying method and traffic sign identifying system based on cascading deep learning |
| CN106326858A (en) * | 2016-08-23 | 2017-01-11 | 北京航空航天大学 | Road traffic sign automatic identification and management system based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| Wei Liu etc..SSD: Single Shot MultiBox Detector.《European Conference on Computer Vision》.2016, |
| 基于图模型与卷积神经网络的交通标志识别方法;刘占文等;《交通运输工程学报》;20161031;第122-130页 |
| 基于联合卷积和递归神经网络的交通标志识别;宣森炎等;《传感器与微系统》;20140331;第33卷(第8期);第30-33页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106909886A (en) | 2017-06-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106909886B (en) | A kind of high-precision method for traffic sign detection and system based on deep learning | |
| CN108037770B (en) | Unmanned aerial vehicle power transmission line inspection system and method based on artificial intelligence | |
| CN111914813B (en) | Power transmission line inspection image naming method and system based on image classification | |
| CN111213155A (en) | Image processing method, device, movable platform, unmanned aerial vehicle and storage medium | |
| CN111582234B (en) | Large-scale oil tea tree forest fruit intelligent detection and counting method based on UAV and deep learning | |
| CN111784685A (en) | Power transmission line defect image identification method based on cloud edge cooperative detection | |
| CN108010030A (en) | A kind of Aerial Images insulator real-time detection method based on deep learning | |
| CN107085696A (en) | A kind of vehicle location and type identifier method based on bayonet socket image | |
| CN108647585A (en) | A kind of traffic mark symbol detection method based on multiple dimensioned cycle attention network | |
| CN107444665A (en) | A kind of unmanned plane Autonomous landing method | |
| CN113160062B (en) | Infrared image target detection method, device, equipment and storage medium | |
| CN108776772A (en) | Across the time building variation detection modeling method of one kind and detection device, method and storage medium | |
| CN108648169A (en) | The method and device of high voltage power transmission tower defects of insulator automatic identification | |
| CN106897681A (en) | A kind of remote sensing images comparative analysis method and system | |
| CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
| CN103679674A (en) | Method and system for splicing images of unmanned aircrafts in real time | |
| CN109815859A (en) | A kind of inferior grade road automatically extracts and mutation analysis method | |
| CN109800712B (en) | Vehicle detection counting method and device based on deep convolutional neural network | |
| CN109426773A (en) | A kind of roads recognition method and device | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 | |
| CN115690081A (en) | Tree counting method, system, storage medium, computer equipment and terminal | |
| CN109657540B (en) | Withered tree positioning method and system | |
| CN109726637A (en) | The parking site detecting system and method in parking stall region are indicated with resilient polygonal | |
| CN112906648A (en) | Method and device for classifying objects in land parcel and electronic equipment | |
| KR102416714B1 (en) | System and method for city-scale tree mapping using 3-channel images and multiple 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 |