CN107122776A - A kind of road traffic sign detection and recognition methods based on convolutional neural networks - Google Patents
A kind of road traffic sign detection and recognition methods based on convolutional neural networks Download PDFInfo
- Publication number
- CN107122776A CN107122776A CN201710243464.9A CN201710243464A CN107122776A CN 107122776 A CN107122776 A CN 107122776A CN 201710243464 A CN201710243464 A CN 201710243464A CN 107122776 A CN107122776 A CN 107122776A
- Authority
- CN
- China
- Prior art keywords
- traffic sign
- layer
- convolutional neural
- neural networks
- pixel
- 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
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- 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
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- 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
- G06V10/56—Extraction of image or video features relating to colour
Abstract
A kind of road traffic sign detection based on convolutional neural networks and recognition methods is claimed in the present invention, belongs to Digital Image Processing and machine learning field.Including step：First, pretreated RGB image is transformed into hsv color space, area-of-interest is obtained by threshold value setting.Then, two classification convolutional neural networks of traffic sign and non-traffic sign are distinguished in design, and whether it is traffic sign that area-of-interest is judged with this.Behind the position for obtaining traffic sign, use the traffic sign recognition method based on convolutional neural networks, the parameters such as the number of plies, the characteristic pattern quantity of convolutional neural networks are adjusted, by a large amount of training samples come the parameter in learning network, and then the classification of the traffic sign of diverse location are recognized.Experiment shows that this method has well adapting to property to the deformation of traffic sign, partial occlusion, inclination etc., and good performance is embodied in terms of recognition effect and recognition efficiency.
Description
Technical field
The invention belongs to Digital Image Processing and machine learning field, and in particular to a kind of based on convolutional neural networks
Traffic sign recognition method.
Background technology
Early in last century early eighties, some countries begin to pay close attention to Traffic Sign Recognition, main using as utilized
The methods such as template matches, rim detection, neutral net.Road traffic sign detection with identification image sources in vehicle picture pick-up device,
Although traffic sign typically all has obvious color and shape feature, because outdoor natural situation is complicated and changeable,
The image of collection is easily influenceed by many unfavorable factors, and such as weather influence, ambient interferences and object block factor, and this can be straight
Connect influence road traffic sign detection and the result of identification.For the detection-phase of traffic sign, Major Difficulties are the change of illumination
So that violent change occurs for traffic sign outward appearance, cause simply to utilize RGB color model to be difficult to its accurate segmentation；Traffic
The cognitive phase of mark, due to target by partial occlusion or because of the pattern distortion of shooting visual angle generation, can cause recognizer
It can decline.To improve road traffic sign detection at this stage and robustness, real-time and the discrimination of identification, this technology is proposed
A kind of road traffic sign detection and recognition methods based on convolutional neural networks.
Not only discrimination reaches 95.46% to this method in the experiment of GTRSB databases, more than existing other methods,
And the time of one width traffic sign of identification only has 0.02s or so, meets real-time.Meanwhile, conventional method is for traffic sign
Identification need it is artificial extract feature, then train grader with the big measure feature of extraction.Set forth herein based on convolutional Neural
The method of network, it is not necessary to artificial to go to extract feature, directly using X-Y scheme as input, is automatically extracted when training
Feature is simultaneously trained, and greatly reduces the plenty of time that feature extraction needs to expend, and effect is more preferable.Moreover, due to volume
The special construction of product neutral net, the well adapting to property such as disturb, blocks on a small quantity, translates on a small quantity for multiple dimensioned.
The content of the invention
Present invention seek to address that above problem of the prior art.Propose one kind greatly reduce feature extraction need consumption
The plenty of time taken, and effect more preferable road traffic sign detection and recognition methods based on convolutional neural networks.The present invention's
Technical scheme is as follows：
A kind of road traffic sign detection and recognition methods based on convolutional neural networks, it comprises the following steps：
101st, the input picture for including traffic sign is obtained, and using RGB of the histogram equalization to the input picture
Image carries out pretreatment operation；
102nd, pretreated RGB image is converted into HSV (hue, saturation, intensity) color model, judges pixel
Whether it is object pixel (i.e. the pixel in traffic sign region), extracts target hsv color information, is carried out just with reference to 8 connected regions
Step segmentation obtains area-of-interest；
103rd, two classification convolutional neural networks of traffic sign and non-traffic sign are distinguished in design, judge that sense is emerging with this
Whether interesting region is traffic sign；
104th, behind the position for obtaining traffic sign, the traffic sign recognition method based on convolutional neural networks, algorithm are used
Parameter of the parameter of setting including the number of plies, characteristic pattern quantity of convolutional neural networks, by training sample come learning network
In parameter, and then recognize diverse location traffic sign classification.
Further, the step 102RGB images are converted to hsv color model and specifically included：Assuming that all colors
Value has all normalized to [0,1], in RGB tri- components Rs, G, B, it is assumed that maximum for MAX, minimum for MIN, then RGB
Be converted to HSV formula as follows：
V=MAX.
Further, it is described to judge whether pixel is that object pixel includes step：
Setpoint color threshold value, is otherwise background pixel if pixel value is object pixel in threshold range, is met
The HSV pixel values set below are red objects pixel
A) the or Hue ＞ 0.95 of Hue ＜ 0.05
B) Saturation ＞ 0.5
C) Value ＞ 0.15.
Further, the connected region of combination 8, which carries out primary segmentation acquisition area-of-interest, includes：Mesh will only be included
Pixel value or the region close with target pixel value are marked as binary map, find has same pixel value and position wherein
The image-region of adjacent foreground pixel point composition, finds 8 connected regions, area-of-interest is obtained by primary segmentation.
Further, the two classification convolutional neural networks one for distinguishing traffic sign and non-traffic sign of the step 103
7 layers are included altogether, first layer is input layer, middle hidden layer is 2 convolutional layers, 2 sample levels and a full articulamentum, most
Later layer is output layer；The input layer is used for input picture matrix；The second layer is the first convolutional layer, is connected with input layer, should
Layer includes 6 characteristic patterns, you can to obtain 6 kinds of different features；Third layer is the first down-sampling layer, is connected with the first convolutional layer
Connect, the layer includes 6 characteristic patterns, calculated and obtained by max-pooling, be also to obtain 6 kinds of different features；4th layer is
Second convolutional layer, be connected, the layer one has 12 characteristic patterns with down-sampling layer, and layer 5 is the second down-sampling layer, and under first
Sample level is substantially similar；Layer 6 is full articulamentum, altogether comprising 196 nodes；Last layer is output layer, and output node is
2, two output neurons of output layer represent traffic sign and non-traffic sign.
Advantages of the present invention and have the beneficial effect that：
1. the present invention utilizes hsv color model extraction feature, the approximate region of traffic sign is determined.
2. the present invention detects the accurate location of traffic sign using two classification convolutional neural networks.
3. present invention discrimination in the experiment of GTRSB databases reaches 95.46%, more than existing other methods, and
And the time of one width traffic sign of identification only has 0.02s or so, meets real-time.
4. the method proposed by the present invention based on convolutional neural networks, it is not necessary to artificial to go to extract feature, directly with two dimension
Image automatically extracts feature when training and trained as input, and greatly reducing feature extraction needs a large amount of of consuming
Time, and effect is more preferable.In addition, this method for it is multiple dimensioned disturb, block on a small quantity, on a small quantity translation etc. have preferably suitable
Ying Xing.
Brief description of the drawings
Fig. 1 is the system architecture diagram that the present invention provides preferred embodiment.
Fig. 2 is the structured flowchart of road traffic sign detection in embodiment.
Fig. 3 is DCNN (Detection Convolutional Neural Network, DCNN) network structure.
Fig. 4 is the structured flowchart of Traffic Sign Recognition in embodiment.
Fig. 5 is RCNN (Recognition Convolutional Neural Network) network structure.
Fig. 6 is DCNN ROC curve figure.
Fig. 7 is that DCNN precision recalls curve map.
Fig. 8 is RCNN 10 mean error figures of iteration during training convolutional neural networks in GTSRB data sets.
Fig. 9 is RCNN 100 mean error figures of iteration during training convolutional neural networks in GTSRB data sets.
Figure 10 is the recognition accuracy block diagram of the invention with existing Traffic Sign Recognition technology.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, detailed
Carefully describe.Described embodiment is only a part of embodiment of the present invention.
The present invention solve above-mentioned technical problem technical scheme be：
Accompanying drawing 1 is the system architecture diagram of the present invention.It is carried out according to following steps：
Step 1, R, G, B triple channel using histogram equalization respectively to input picture do histogram equalization
Step 2, the RGB image that step 1 is pre-processed is converted into hsv color model, then extracts target color information,
Primary segmentation, which is carried out, with reference to 8 connected regions obtains area-of-interest (Region Of Interest, ROI)
Step 201, RGB image are converted to hsv color model.Assuming that all color values all normalized to [0,
1], minimum for MIN in RGB tri- components Rs, G, B, it is assumed that maximum for MAX, then to be converted to HSV formula as follows by RGB：
V=MAX
Step 202, setpoint color threshold value.It is otherwise background picture if pixel value is object pixel in threshold range
Element.The HSV pixel values for meeting following setting are red objects pixel.
A) the or Hue ＞ 0.95 of Hue ＜ 0.05
B) Saturation ＞ 0.5
C) Value ＞ 0.15
Step 203, only include target pixel value or the region close with target pixel value, as binary map, wherein
The image-region with the adjacent foreground pixel point composition of same pixel value and position is found, 8 connected regions are found, by first
Step segmentation obtains ROIs.
Step 3, classify designed for the two of the road traffic sign detection stage convolutional neural networks, referred to as DCNN
(Detection Convolutional Neural Network).DCNN network structure is as shown in figure 3, the convolutional Neural
Network includes altogether 7 layers, and first layer is input layer, and middle hidden layer is 2 convolutional layers, 2 sample levels and a full connection
Layer, last layer is output layer.
Step 302, for input layer, it is necessary to input data be 48 × 48 pixel sizes image array.Due to from figure
The ROI resolution sizes split as in differ, and ignore the length and width size ratio of input picture in itself, directly the image contracts
It is put into 48 × 48 sizes.
Step 303, the DCNN second layer are convolutional layer, are connected with input layer.The layer includes 6 characteristic patterns, can both obtain
To 6 kinds of different features.The size of convolution kernel is 5 × 5, and the moving step length of convolution kernel is defaulted as 1, therefore, with 5 × 5 convolution
Core goes to do convolution to 48 × 48 sized images of input layer, and obtained characteristic pattern size is (48-5+1) × (48-5+1), is
44×44.Each convolution kernel includes 5 × 5 totally 25 training parameters, an additional bias term, and a total of 6 convolution kernels, institute
So that, it is necessary to which the number of parameters of training is (5 × 5+1) × 6=156, total connection number is 156 × (44 × 44)=302016.
Step 304, DCNN third layer are down-sampling layer, are connected with last layer convolutional layer.The layer includes 6 characteristic patterns,
Calculated and obtained by max-pooling, be also to obtain 6 kinds of different features.The sample range of down-sampling layer is 2 × 2, because
Do not include lap between sampling, therefore moving step length is 2.2 × 2 i.e. 4 pixels to image are sampled every time, therefore
The resolution ratio of obtained characteristic pattern is the 1/4 of last layer, therefore, and the characteristic pattern size of this layer is 22 × 22.Adopted at intervals of 2
Sample calculating process is similar to convolutional calculation process, the difference is that being without overlapping between each zoning.
Step 305, the 4th layer of DCNN be convolutional layer, be connected with last layer down-sampling layer.The layer one has 12 features
Figure, convolution kernel size is 5 × 5, and characteristic pattern resolution ratio is (22-5+1) × (22-5+1), as 18 × 18.The layer and the second layer
Similar, difference is that the input data of the second layer is an image for coming from input layer, different convolution kernel respectively with
It carries out convolution algorithm, and the input data of this layer is not input layer, but the 6 of third layer characteristic patterns are, it is necessary to use this 6
Characteristic pattern obtains 12 characteristic patterns of this layer by calculating.12 kinds of permutation and combination methods of 6 characteristic patterns.One is represented per a line
Characteristic pattern is planted, altogether 6 kinds of characteristic patterns.Each row represent that several characteristic patterns with choosing carry out the spy that convolution algorithm is obtained
Levy figure.By taking this layer of characteristic pattern 1 as an example, this feature figure by down-sampling layer the 2nd, 3,4 characteristic patterns obtain, rolled up using identical
Product checks this corresponding same position of 3 characteristic patterns and does convolution algorithm respectively, and obtain 3 convolution results are passed through
Sigmoid Function Mappings are 1 convolution results, are used as a pixel value of characteristic pattern 1.This combination, because special
The incomplete connection of figure is levied, the quantity of connection can be effectively reduced.
Step 306, DCNN layer 5 are down-sampling layer, substantially similar to third layer.Sample range is 2 × 2, sampling
At intervals of 2, each characteristic pattern size is 9 × 9, totally 12 characteristic patterns, is corresponded with the characteristic pattern of last layer.
Step 307, DCNN layer 6 are full articulamentum, altogether comprising 196 nodes, each node and last layer
Each node connection, therefore the number of weight connection is (9 × 9 × 12) × 196=190512.
Step 308, DCNN last layer are output layer, and output node is 2, is by all nerves of layer second from the bottom
Member obtains probability distribution of the sample respectively in two classes, then according to the classification of high probability as 2 road softmax input
Which kind of is judged as.Because main functions of the DCNN in road traffic sign detection is to judge whether ROI is traffic sign, therefore
Two output neurons of output layer represent traffic sign and non-traffic sign.
Whether step 4, the ROIs for having trained DCNN judgment steps 2 to obtain using step 3 are traffic sign, are to retain,
Otherwise give up, obtain final detection result.
Step 5, the convolutional neural networks designed for the Traffic Sign Recognition stage, referred to as RCNN (Recognition
CNN).RCNN structure as shown in figure 5, contain 9 layers altogether, wherein in the middle of hidden layer be 3 convolutional layers, 3 sample levels and
One full articulamentum.
The image that step 501, RCNN input layer are 48 × 48, as DCNN, input sample and DCNN stages one
Input picture, it is necessary to be normalized to formed objects, i.e. 48 × 48 pixel sizes by sample.The RCNN second layer is convolutional layer, is had
6 characteristic patterns.Convolution kernel size be 77, therefore each characteristic pattern size of the layer be (48-7+1) × (48-7+1), as 42 ×
42.This layer essentially identical with the second layer in DCNN.RCNN third layer is down-sampling layer, uses max-pooling methods pair
The characteristic pattern of a upper convolutional layer carries out sampled operational, and sample range is 2 × 2 sizes, and the sampling interval is 2, obtains 6 sizes and is
21 × 21 characteristic pattern.The 4th layer of RCNN is convolutional layer, totally 16 characteristic patterns, and convolution kernel size is 4 × 4.The layer with
DCNN structure is basically identical, and the mode for obtaining characteristic pattern is also identical with DCNN.RCNN layer 5 is down-sampling layer, with the
Three layers similar.Likewise, layer 6 and layer 7 to it is above similar.Last layer of RCNN is output layer, and output neuron is 43
It is individual, 43 classifications of traffic sign are represented respectively.
Step 6, the traffic sign to step 4 reservation are identified, and obtain the classification of traffic sign.
In order to verify the effect of the present invention, following experiment has been carried out：
1.DCNN is tested in GTSDB data sets
1.1 experimental analysis：
The purpose of road traffic sign detection is to try to remove unnecessary background information, retains the region containing only traffic sign.Institute
So that the convolutional neural networks of road traffic sign detection will not only ensure discrimination, while also to ensure the recall rate of sample.
For two classification problems, four kinds of situations can be obtained：
The negative class FN of vacation：False Negative, are judged as negative sample, but are in fact positive samples.
False positive class FP：False Positive, are judged as positive sample, but are in fact negative samples.
Really bear class TN：True Negative, are judged as negative sample, are in fact also negative sample.
Real class TP：True Positive, are judged as positive sample, are in fact also positive sample.
Wherein TP+FP+TN+FN=total sample numbers.
Following concept is introduced to evaluate two classification convolutional neural networks.
Accuracy (Precision)：
Precision=TP/ (TP+FP)
Real class rate (True Positive Rate, TPR), also referred to as recall rate (Recall)：
TPR=Recall=TP/ (TP+FN)
False positive class rate (False Positive Rate, FPR), also referred to as metastatic (Specificity)：
FPR=FP/ (FP+TN)
Fig. 6 is DCNN ROC curve, and the abscissa of ROC curve is negative and positive class rate (FPR), and ordinate is real class rate
(TPR).In ROC curve, (0,1), i.e. FPR=0, TPR=1, it is meant that FN=0, and FP=0, i.e. classifying quality are special
It is good, without classification error；(1,0), i.e. FPR=1, TPR=0, it is meant that TN=0, and TP=0, i.e. classifying quality are special
Difference, all classification mistake：(0,0), i.e. FPR=TPR=0, it is meant that FP=TP=0, i.e., be all identified as negative sample；(1,
1), i.e. FPR=1, TPR=1, it is meant that FN=TN=0, i.e., be all identified as positive sample.Can from above-mentioned 4 points and image
Go out, point of the point on ROC curve on (0,1), i.e. image is closer to the upper left corner, and classifying quality is better.Fig. 6 is two points
The ROC curve of class convolutional neural networks, the curve closely upper left corner, so the classification of the two classification convolutional neural networks
Effect is fine.Fig. 7 is that DCNN precision recalls curve, and DCNN discrimination is 98%, and corresponding recall rate is 97.8%, substantially
Meet discrimination and all compared high requirement with recall rate.
2.RCNN is tested in GTSRB data sets
2.1 experimental analysis：
During an iteration, whole training sample is divided into several pieces, per a data for containing identical quantity,
Such as 50.50 traffic indication maps are inputted when training convolutional neural networks each time, this 50 figures are obtained by propagated forward
The error of picture, then carries out derivation, and backpropagation undated parameter according to this error to network weight and biasing.I.e. once
With 50 images come undated parameter.
Assuming that the number of times of parameter adjustment is n, then n=number of training/50 × iterations.Fig. 8, Fig. 9 are respectively illustrated
Convolutional neural networks distinguish the square error rate convergence curve that undated parameter number of times is 7840,78400 on training dataset.
I.e. training network difference iteration 10,100 times.Observation finds to get over 50,000 times in parameter adjustment, i.e., iteration 64 times or so when square mistake
Difference tends to 0.01, and is held essentially constant.
Figure 10 is to use GTSRB data sets, by extracting HOG, LBP, Haar feature combination SVM sorting technique, Yi Jili
Compared with LRC (linear regression) method and context of methods the recognition correct rates classified.
The above embodiment is interpreted as being merely to illustrate the present invention rather than limited the scope of the invention.
After the content of record of the present invention has been read, technical staff can make various changes or modifications to the present invention, and these are equivalent
Change and modification equally fall into the scope of the claims in the present invention.
Claims (5)
1. a kind of road traffic sign detection and recognition methods based on convolutional neural networks, it is characterised in that comprise the following steps：
101st, the input picture for including traffic sign is obtained, and using RGB image of the histogram equalization to the input picture
Carry out pretreatment operation；
102nd, pretreated RGB image is converted into hsv color model, H, S, V represent hue, saturation, intensity respectively, sentence
Whether disconnected pixel is the object pixel i.e. pixel in traffic sign region, extracts object pixel colouring information, enters with reference to 8 connected regions
Row primary segmentation obtains area-of-interest；
103rd, two classification convolutional neural networks of traffic sign and non-traffic sign are distinguished in design, and area-of-interest is judged with this
Whether it is traffic sign；
104th, behind the position for obtaining traffic sign, using the traffic sign recognition method based on convolutional neural networks, algorithm is set
Parameter of the parameter including the number of plies, characteristic pattern quantity of convolutional neural networks, by training sample come in learning network
Parameter, and then recognize the classification of the traffic sign of diverse location.
2. road traffic sign detection and recognition methods according to claim 1 based on convolutional neural networks, it is characterised in that
The step 102RGB images are converted to hsv color model and specifically included：Assuming that all color values all normalized to [0,
1], minimum for MIN in RGB tri- components Rs, G, B, it is assumed that maximum for MAX, then to be converted to HSV formula as follows by RGB：
<mrow>
<mi>S</mi>
<mo>=</mo>
<mfrac>
<mrow>
<mi>M</mi>
<mi>A</mi>
<mi>X</mi>
<mo>-</mo>
<mi>M</mi>
<mi>I</mi>
<mi>N</mi>
</mrow>
<mrow>
<mi>M</mi>
<mi>A</mi>
<mi>X</mi>
</mrow>
</mfrac>
</mrow>
V=MAX.
3. road traffic sign detection and recognition methods according to claim 2 based on convolutional neural networks, it is characterised in that
It is described to judge whether pixel is that object pixel includes step：
Setpoint color threshold value, is otherwise background pixel if pixel value is object pixel in threshold range, meets to divide into
Fixed HSV pixel values are red objects pixel
A) Hue ＜ 0.05orHue ＞ 0.95
B) Saturation ＞ 0.5
C) Value ＞ 0.15.
4. road traffic sign detection and recognition methods according to claim 3 based on convolutional neural networks, it is characterised in that
The connected region of combination 8, which carries out primary segmentation acquisition area-of-interest, to be included：Target pixel value or and target will only be included
Found wherein with the adjacent foreground pixel point composition of same pixel value and position as binary map in the close region of pixel value
Image-region, find 8 connected regions, by primary segmentation obtain area-of-interest.
5. road traffic sign detection based on convolutional neural networks and recognition methods according to one of claim 1-4, it is special
Levy and be, the two classification convolutional neural networks for distinguishing traffic sign and non-traffic sign of the step 103 include altogether 7 layers,
First layer is input layer, and middle hidden layer is 2 convolutional layers, 2 sample levels and a full articulamentum, and last layer is output
Layer；The input layer is used for input picture matrix；The second layer is the first convolutional layer, is connected with input layer, and the layer includes 6 features
Figure, you can to obtain 6 kinds of different features；Third layer is the first down-sampling layer, is connected with the first convolutional layer, and the layer includes 6
Characteristic pattern, is calculated by max-pooling and obtained, be also to obtain 6 kinds of different features；4th layer is the second convolutional layer, with
Sample level is connected, and the layer one has 12 characteristic patterns, and layer 5 is the second down-sampling layer, substantially similar to the first down-sampling layer；
Layer 6 is full articulamentum, altogether comprising 196 nodes；Last layer is output layer, and output node is 2, and two of output layer are defeated
Go out neuron and represent traffic sign and non-traffic sign.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710243464.9A CN107122776A (en) | 2017-04-14 | 2017-04-14 | A kind of road traffic sign detection and recognition methods based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710243464.9A CN107122776A (en) | 2017-04-14 | 2017-04-14 | A kind of road traffic sign detection and recognition methods based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107122776A true CN107122776A (en) | 2017-09-01 |
Family
ID=59725634
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710243464.9A Pending CN107122776A (en) | 2017-04-14 | 2017-04-14 | A kind of road traffic sign detection and recognition methods based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107122776A (en) |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107463927A (en) * | 2017-09-21 | 2017-12-12 | 广东工业大学 | A kind of deceleration driven detection method and device based on convolutional neural networks |
| CN107644221A (en) * | 2017-10-31 | 2018-01-30 | 南京航空航天大学 | Convolutional neural networks traffic sign recognition method based on compression of parameters |
| CN108154102A (en) * | 2017-12-21 | 2018-06-12 | 安徽师范大学 | A kind of traffic sign recognition method |
| CN108154502A (en) * | 2017-12-22 | 2018-06-12 | 王华锋 | A kind of though-hole solder joint recognition methods based on convolutional neural networks |
| CN108304785A (en) * | 2018-01-16 | 2018-07-20 | 桂林电子科技大学 | Road traffic sign detection based on self-built neural network and recognition methods |
| CN108416270A (en) * | 2018-02-06 | 2018-08-17 | 南京信息工程大学 | A kind of traffic sign recognition method based on more attribute union features |
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108564048A (en) * | 2018-04-20 | 2018-09-21 | 天津工业大学 | A kind of depth convolutional neural networks method applied to Traffic Sign Recognition |
| CN108710826A (en) * | 2018-04-13 | 2018-10-26 | 燕山大学 | A kind of traffic sign deep learning mode identification method |
| CN108898078A (en) * | 2018-06-15 | 2018-11-27 | 上海理工大学 | A kind of traffic sign real-time detection recognition methods of multiple dimensioned deconvolution neural network |
| CN109446984A (en) * | 2018-10-26 | 2019-03-08 | 重庆大学 | Traffic sign recognition method in natural scene |
| CN109492526A (en) * | 2018-09-27 | 2019-03-19 | 桂林电子科技大学 | Traffic sign recognition method based on LDCNN model and NHE algorithm |
| CN109726728A (en) * | 2017-10-31 | 2019-05-07 | 高德软件有限公司 | A kind of training data generation method and device |
| CN109815906A (en) * | 2019-01-25 | 2019-05-28 | 华中科技大学 | Method for traffic sign detection and system based on substep deep learning |
| CN109886325A (en) * | 2019-02-01 | 2019-06-14 | 辽宁工程技术大学 | A kind of stencil-chosen and acceleration matching process of non linear color space classification |
| CN109919002A (en) * | 2019-01-23 | 2019-06-21 | 平安科技（深圳）有限公司 | The recognition methods of yellow stop forbidden line, device, computer equipment and storage medium |
| CN110163187A (en) * | 2019-06-02 | 2019-08-23 | 东北石油大学 | Remote road traffic sign detection recognition methods based on F-RCNN |
| CN110200601A (en) * | 2019-06-17 | 2019-09-06 | 广东工业大学 | A kind of pulse condition acquisition device and system |
| CN110276760A (en) * | 2018-03-14 | 2019-09-24 | 广东欧珀移动通信有限公司 | A kind of image scene segmentation method, terminal and storage medium |
| CN110390228A (en) * | 2018-04-20 | 2019-10-29 | 北京四维图新科技股份有限公司 | The recognition methods of traffic sign picture, device and storage medium neural network based |
| CN110633635A (en) * | 2019-08-08 | 2019-12-31 | 北京联合大学 | ROI-based traffic sign board real-time detection method and system |
| CN110909756A (en) * | 2018-09-18 | 2020-03-24 | 苏宁 | Convolutional neural network model training method and device for medical image recognition |
| CN111242046A (en) * | 2020-01-15 | 2020-06-05 | 江苏北斗星通汽车电子有限公司 | Ground traffic sign identification method based on image retrieval |
| CN111259886A (en) * | 2020-01-08 | 2020-06-09 | 上海眼控科技股份有限公司 | License plate screw detection method, electronic device, computer equipment and storage medium |
| CN111274971A (en) * | 2020-01-21 | 2020-06-12 | 南京航空航天大学 | Traffic identification method based on color space fusion network and space transformation network |
| CN111428556A (en) * | 2020-02-17 | 2020-07-17 | 浙江树人学院（浙江树人大学） | Traffic sign recognition method based on capsule neural network |
| CN112101372A (en) * | 2019-06-17 | 2020-12-18 | 北京地平线机器人技术研发有限公司 | Method and device for determining image characteristics |
| CN112200200A (en) * | 2020-10-12 | 2021-01-08 | 蚌埠依爱消防电子有限责任公司 | LED light color detection method |
| CN113536943A (en) * | 2021-06-21 | 2021-10-22 | 上海赫千电子科技有限公司 | Road traffic sign identification method based on image enhancement |
| CN117576492A (en) * | 2024-01-18 | 2024-02-20 | 天津医科大学第二医院 | Automatic focus marking and identifying device for gastric interstitial tumor under gastric ultrasonic endoscope |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184393A (en) * | 2011-06-20 | 2011-09-14 | 苏州两江科技有限公司 | Method for judging automobile type according to license plate recognition |
| CN104850845A (en) * | 2015-05-30 | 2015-08-19 | 大连理工大学 | Traffic sign recognition method based on asymmetric convolution neural network |
| CN105488797A (en) * | 2015-11-25 | 2016-04-13 | 安徽创世科技有限公司 | License plate location method for HSV space |
| CN106372571A (en) * | 2016-08-18 | 2017-02-01 | 宁波傲视智绘光电科技有限公司 | Road traffic sign detection and identification method |
-
2017
- 2017-04-14 CN CN201710243464.9A patent/CN107122776A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184393A (en) * | 2011-06-20 | 2011-09-14 | 苏州两江科技有限公司 | Method for judging automobile type according to license plate recognition |
| CN104850845A (en) * | 2015-05-30 | 2015-08-19 | 大连理工大学 | Traffic sign recognition method based on asymmetric convolution neural network |
| CN105488797A (en) * | 2015-11-25 | 2016-04-13 | 安徽创世科技有限公司 | License plate location method for HSV space |
| CN106372571A (en) * | 2016-08-18 | 2017-02-01 | 宁波傲视智绘光电科技有限公司 | Road traffic sign detection and identification method |
Non-Patent Citations (5)
| Title |
|---|
| 庄家俊等: "面向辅助驾驶的夜间行人检测方法", 《华南理工大学学报（自然科学版）》 * |
| 张金朋等: "融合颜色分割与形状特征的交通标志检测", 《微型机与应用》 * |
| 李伦波: "自然场景下交通标志的检测与分类算法研究", 《万方数据》 * |
| 谭贞刚: "基于多分辨率卷积网络的交通标志检测与识别", 《中国优秀硕士学位论文全文数据库工程科技II辑》 * |
| 赵谦等: "《智能视频图像处理技术与应用》", 30 September 2016 * |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107463927A (en) * | 2017-09-21 | 2017-12-12 | 广东工业大学 | A kind of deceleration driven detection method and device based on convolutional neural networks |
| CN107644221A (en) * | 2017-10-31 | 2018-01-30 | 南京航空航天大学 | Convolutional neural networks traffic sign recognition method based on compression of parameters |
| CN109726728A (en) * | 2017-10-31 | 2019-05-07 | 高德软件有限公司 | A kind of training data generation method and device |
| CN108154102B (en) * | 2017-12-21 | 2021-12-10 | 安徽师范大学 | Road traffic sign identification method |
| CN108154102A (en) * | 2017-12-21 | 2018-06-12 | 安徽师范大学 | A kind of traffic sign recognition method |
| CN108154502A (en) * | 2017-12-22 | 2018-06-12 | 王华锋 | A kind of though-hole solder joint recognition methods based on convolutional neural networks |
| CN108304785A (en) * | 2018-01-16 | 2018-07-20 | 桂林电子科技大学 | Road traffic sign detection based on self-built neural network and recognition methods |
| CN108416270A (en) * | 2018-02-06 | 2018-08-17 | 南京信息工程大学 | A kind of traffic sign recognition method based on more attribute union features |
| CN108416270B (en) * | 2018-02-06 | 2021-07-06 | 南京信息工程大学 | Traffic sign identification method based on multi-attribute combined characteristics |
| CN110276760B (en) * | 2018-03-14 | 2020-11-17 | Oppo 广东移动通信有限公司 | Image scene segmentation method, terminal and storage medium |
| CN110276760A (en) * | 2018-03-14 | 2019-09-24 | 广东欧珀移动通信有限公司 | A kind of image scene segmentation method, terminal and storage medium |
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108710826A (en) * | 2018-04-13 | 2018-10-26 | 燕山大学 | A kind of traffic sign deep learning mode identification method |
| CN108564048A (en) * | 2018-04-20 | 2018-09-21 | 天津工业大学 | A kind of depth convolutional neural networks method applied to Traffic Sign Recognition |
| CN110390228B (en) * | 2018-04-20 | 2022-05-10 | 北京四维图新科技股份有限公司 | Traffic sign picture identification method and device based on neural network and storage medium |
| CN110390228A (en) * | 2018-04-20 | 2019-10-29 | 北京四维图新科技股份有限公司 | The recognition methods of traffic sign picture, device and storage medium neural network based |
| CN108898078A (en) * | 2018-06-15 | 2018-11-27 | 上海理工大学 | A kind of traffic sign real-time detection recognition methods of multiple dimensioned deconvolution neural network |
| CN110909756A (en) * | 2018-09-18 | 2020-03-24 | 苏宁 | Convolutional neural network model training method and device for medical image recognition |
| CN109492526A (en) * | 2018-09-27 | 2019-03-19 | 桂林电子科技大学 | Traffic sign recognition method based on LDCNN model and NHE algorithm |
| CN109446984A (en) * | 2018-10-26 | 2019-03-08 | 重庆大学 | Traffic sign recognition method in natural scene |
| CN109919002A (en) * | 2019-01-23 | 2019-06-21 | 平安科技（深圳）有限公司 | The recognition methods of yellow stop forbidden line, device, computer equipment and storage medium |
| CN109919002B (en) * | 2019-01-23 | 2024-02-27 | 平安科技（深圳）有限公司 | Yellow stop line identification method and device, computer equipment and storage medium |
| CN109815906A (en) * | 2019-01-25 | 2019-05-28 | 华中科技大学 | Method for traffic sign detection and system based on substep deep learning |
| CN109886325A (en) * | 2019-02-01 | 2019-06-14 | 辽宁工程技术大学 | A kind of stencil-chosen and acceleration matching process of non linear color space classification |
| CN109886325B (en) * | 2019-02-01 | 2022-11-29 | 辽宁工程技术大学 | Template selection and accelerated matching method for nonlinear color space classification |
| CN110163187A (en) * | 2019-06-02 | 2019-08-23 | 东北石油大学 | Remote road traffic sign detection recognition methods based on F-RCNN |
| CN110200601A (en) * | 2019-06-17 | 2019-09-06 | 广东工业大学 | A kind of pulse condition acquisition device and system |
| CN110200601B (en) * | 2019-06-17 | 2022-04-19 | 广东工业大学 | Pulse condition acquisition device and system |
| CN112101372A (en) * | 2019-06-17 | 2020-12-18 | 北京地平线机器人技术研发有限公司 | Method and device for determining image characteristics |
| CN110633635A (en) * | 2019-08-08 | 2019-12-31 | 北京联合大学 | ROI-based traffic sign board real-time detection method and system |
| CN111259886A (en) * | 2020-01-08 | 2020-06-09 | 上海眼控科技股份有限公司 | License plate screw detection method, electronic device, computer equipment and storage medium |
| CN111242046B (en) * | 2020-01-15 | 2023-08-29 | 江苏北斗星通汽车电子有限公司 | Ground traffic sign recognition method based on image retrieval |
| CN111242046A (en) * | 2020-01-15 | 2020-06-05 | 江苏北斗星通汽车电子有限公司 | Ground traffic sign identification method based on image retrieval |
| CN111274971A (en) * | 2020-01-21 | 2020-06-12 | 南京航空航天大学 | Traffic identification method based on color space fusion network and space transformation network |
| CN111428556A (en) * | 2020-02-17 | 2020-07-17 | 浙江树人学院（浙江树人大学） | Traffic sign recognition method based on capsule neural network |
| CN111428556B (en) * | 2020-02-17 | 2024-01-19 | 浙江树人学院（浙江树人大学） | Traffic sign recognition method based on capsule neural network |
| CN112200200A (en) * | 2020-10-12 | 2021-01-08 | 蚌埠依爱消防电子有限责任公司 | LED light color detection method |
| CN112200200B (en) * | 2020-10-12 | 2023-04-18 | 蚌埠依爱消防电子有限责任公司 | LED light color detection method |
| CN113536943A (en) * | 2021-06-21 | 2021-10-22 | 上海赫千电子科技有限公司 | Road traffic sign identification method based on image enhancement |
| CN113536943B (en) * | 2021-06-21 | 2024-04-12 | 上海赫千电子科技有限公司 | Road traffic sign recognition method based on image enhancement |
| CN117576492A (en) * | 2024-01-18 | 2024-02-20 | 天津医科大学第二医院 | Automatic focus marking and identifying device for gastric interstitial tumor under gastric ultrasonic endoscope |
| CN117576492B (en) * | 2024-01-18 | 2024-03-29 | 天津医科大学第二医院 | Automatic focus marking and identifying device for gastric interstitial tumor under gastric ultrasonic endoscope |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107122776A (en) | A kind of road traffic sign detection and recognition methods based on convolutional neural networks | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN109284669A (en) | Pedestrian detection method based on Mask RCNN | |
| CN105512684B (en) | Logo automatic identifying method based on principal component analysis convolutional neural networks | |
| CN108154102B (en) | Road traffic sign identification method | |
| CN104598885B (en) | The detection of word label and localization method in street view image | |
| CN103514456B (en) | Image classification method and device based on compressed sensing multi-core learning | |
| CN109684922B (en) | Multi-model finished dish identification method based on convolutional neural network | |
| CN108108761A (en) | A kind of rapid transit signal lamp detection method based on depth characteristic study | |
| CN105809121A (en) | Multi-characteristic synergic traffic sign detection and identification method | |
| CN108734189A (en) | Vehicle License Plate Recognition System based on atmospherical scattering model and deep learning under thick fog weather | |
| CN106650786A (en) | Image recognition method based on multi-column convolutional neural network fuzzy evaluation | |
| CN107103317A (en) | Fuzzy license plate image recognition algorithm based on image co-registration and blind deconvolution | |
| CN104484681B (en) | Hyperspectral Remote Sensing Imagery Classification method based on spatial information and integrated study | |
| CN106919920A (en) | Scene recognition method based on convolution feature and spatial vision bag of words | |
| CN108009518A (en) | A kind of stratification traffic mark recognition methods based on quick two points of convolutional neural networks | |
| CN107392130A (en) | Classification of Multispectral Images method based on threshold adaptive and convolutional neural networks | |
| CN111428556B (en) | Traffic sign recognition method based on capsule neural network | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN104778453A (en) | Night pedestrian detection method based on statistical features of infrared pedestrian brightness | |
| CN105955708A (en) | Sports video lens classification method based on deep convolutional neural networks | |
| CN108647700A (en) | Multitask vehicle part identification model based on deep learning, method and system | |
| CN105718866A (en) | Visual target detection and identification method | |
| CN106022254A (en) | Image recognition technology | |
| CN106503748A (en) | A kind of based on S SIFT features and the vehicle targets of SVM training aids |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170901 |