CN109949593A - A kind of traffic lights recognition methods and system based on crossing priori knowledge - Google Patents
A kind of traffic lights recognition methods and system based on crossing priori knowledge Download PDFInfo
- Publication number
- CN109949593A CN109949593A CN201910189015.XA CN201910189015A CN109949593A CN 109949593 A CN109949593 A CN 109949593A CN 201910189015 A CN201910189015 A CN 201910189015A CN 109949593 A CN109949593 A CN 109949593A
- Authority
- CN
- China
- Prior art keywords
- traffic lights
- crossing
- image
- lane
- priori knowledge
- 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
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention provides a kind of traffic lights recognition methods and system based on crossing priori knowledge, and wherein method includes input traffic scene image, further comprising the steps of: according to the offline labeled data of acquisition, to construct the traffic lights model of specific region；Identification of the training convolutional neural networks to traffic lights；Export recognition result.The present invention can be quickly obtained the candidate region of traffic lights using the positioning of the prior information additional transport signal lamp at crossing, and depth model is then recycled to carry out the identification of traffic lights.It is fast with processing speed, the high feature of accuracy.
Description
Technical field
The present invention relates to the technical field of image vision, especially a kind of traffic lights based on crossing priori knowledge are known
Other method and system.
Background technique
Road environment perception is an important factor for influencing vehicle drive, and wherein traffic lights can make vehicle in four crossway
Mouthful orderly, safe passing, increase substantially the traffic efficiency at crossing.Therefore, the signal lamp identifying system of precise and high efficiency is intelligence
Vehicle environmental perceives indispensable component part.Domestic and foreign scholars have been carried out the research of many years traffic lights identification, take
Obtain much progress and achievement.
The recognition methods of existing traffic lights is broadly divided into the method based on image procossing and is passed based on the network information
Defeated two class of method.Method based on image procossing usually use camera acquisition vehicle front video image, then according to
By the infomation detections such as color, shape area-of-interest and feature progress kind judging is extracted, to obtain the shape of traffic lights
Condition.According to the characteristic attribute of traffic lights, the detection of signal lamp can be divided into detection method based on color space, based on shape
The detection method of shape feature and some other comprehensive method.Algorithm real-time based on color space detection is preferable, is mesh
More one of method is used in preceding traffic lights detection method.Such method is strong for environmental Comparison, color is apparent
Image can obtain relatively good as a result, however only can not cope with the signal lamp detection under complex background by colouring information.Shape
Shape detection method can overcome the influence for the problems such as color is fuzzy, uneven illumination is even, if but that there are shapes under complex background is similar
Chaff interferent then will lead to the failure of signal lamp detection.Separately there are some research methods to comprehensively consider color and shape.Compared to list
Certain feature is solely utilized, such mode can be further reduced identification error rate, but still can not when night, bad weather
Reach the real-time and robustness requirement of system.
The patent of invention of Publication No. CN108764216A discloses a kind of traffic lights recognition methods of view-based access control model
And device, wherein recognition methods includes three steps, and the first step determines traffic signals lamp type, second step image preprocessing, third step
Deep learning identification model.The traffic lights recognition methods referred in the patent, the image of input are traffic lights parts
Area image, then by threshold decision be what kind of signal lamp (horizontal, vertical), then empirical value is split, finally again
The identification of traffic signals classification is carried out to the image after segmentation.This method depends on empirical value, is difficult real in practical applications
It applies, and this method does not refer to how carrying out positioning this committed step to traffic lights region, further increases and know to last
The uncertainty of other result.
Summary of the invention
In order to solve the above technical problems, the present invention proposes a kind of traffic lights identification based on crossing priori knowledge
Method and system, the present invention can be quickly obtained traffic signals using the positioning of the prior information additional transport signal lamp at crossing
Then the candidate region of lamp recycles depth model to carry out the identification of traffic lights.Fast with processing speed, accuracy is high
Feature.
The first object of the present invention is to provide a kind of traffic lights recognition methods based on crossing priori knowledge, including defeated
Enter traffic scene image, further comprising the steps of:
Step 1: according to the offline labeled data of acquisition, constructing the traffic lights model of specific region；
Step 2: identification of the training convolutional neural networks to traffic lights；
Step 3: output recognition result.
Preferably, the offline labeled data acquisition method the following steps are included:
Step 01: since apart from crossing N meters, vehicle starts to acquire GPS data and image data in lane, wherein N
For natural number；
Step 02: acquiring a frame image every Tms and record the differential GPS point coordinate data and vehicle at Image Acquisition moment
Road information, wherein T is natural number；
Step 03: from the road rightmost side, car lane is numbered to the left, lane number information where record current vehicle.
In any of the above-described scheme preferably, it acquires the offline labeled data and needs to meet following require:
1) the offline labeled data in each lane is acquired；
2) in collection process, vehicle driving need to lane center and travel speed be less than or equal to threshold speed.
In any of the above-described scheme preferably, the mask method of the offline labeled data the following steps are included:
Step 11: according to the collected data, every frame image being labeled, marks out image with rectangle frame in the picture
In area-of-interest；
Step 12: the GPS point coordinate information at the rectangle frame coordinate of mark acquisition moment corresponding with image is integrated into one 7
Tuple, [1., 2., 3., 4., 5., 6., 7.].
In any of the above-described scheme preferably, the interested region is rectangle frame, including four coordinate points difference
Upper left, lower-left, upper right and the lower-right most point coordinate of corresponding rectangle frame.
In any of the above-described scheme preferably, the rectangle frame need to cover the traffic lights region in image, and
Edge region has certain surplus space.
In any of the above-described scheme preferably, in 7 tuple, 1. and 2. for storing latitude and longitude coordinates point, 3.
For storing lane number information, it 4., 5., 6. and to be 7. used for storage rectangle frame coordinate.
In any of the above-described scheme preferably, the storage format of the offline labeled data is each crossing according to motor-driven
Lane quantity generates the storage file named with lane information, and every a line correspondence in the storage file stores described 7 yuan
Group data represent the area-of-interest position in corresponding lane in vehicle GPS coordinate downward view.
In any of the above-described scheme preferably, the application method of the offline labeled data is when intelligent vehicle is close to road
When mouth, according to the GPS information that vehicle positioning system positions, nearest point, acquisition pair are matched in acquired off-line data
The rectangle frame position for the area-of-interest answered.
In any of the above-described scheme preferably, the color conversion formula in the traffic lights model are as follows:
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', G ', B '), Cmin=min (R ', G ',
B '), Δ=Cmax-Cmin, the normalized red component of R ' expression, R expression red component, the normalized green component of G ' expression, G
Indicate that green component, B ' indicate that normalized blue component, B indicate blue component, CmaxIndicate R ', G ', B after normalizing ' most
Big person, CminIndicate that the reckling of R ', G ', B after normalizing ', Δ indicate the difference of the maximum and reckling.
In any of the above-described scheme preferably, the step 2 includes red, green according to the traffic lights model extraction
With yellow color region.
In any of the above-described scheme preferably, the recognition methods of the signal lamp the following steps are included:
Step 21: obtaining corresponding traffic signals region using GPS positioning；
Step 22: being detected in the traffic signals region using sliding window method, different sliding window sizes
It can obtain different rectangle frames；
Step 23: being screened using the rectangle frame of the non-maximum restraining method to overlapping
In any of the above-described scheme preferably, the non-maximum restraining method the following steps are included:
Step 231: in all bounding box outputs found out, rejecting the bounding box of pc≤0.6, pc is indicated ...；
Step 232: selecting the corresponding bounding box of maximum pc at this time, export the bounding box, and delete the bounding box；
Step 233: the corresponding bounding box of maximum pc at this time is selected again, and the calculating of loU is carried out with the bounding box of output,
If result >=0.5 of loU, not output boundary frame；Otherwise output boundary frame, and delete the bounding box；
Step 234: step 233 is repeated, until having handled all bounding boxes.
The second object of the present invention is to provide a kind of traffic lights identifying system based on crossing priori knowledge, including takes the photograph
Further include with lower module as head and image input module:
Signal identification module: for the image according to input, traffic lights are identified.
Output module: output recognition result；
The system is implemented according to the method for claim 1.
Preferably, the signal identification module is used for the offline labeled data according to acquisition, constructs the friendship of specific region
Ventilating signal lamp model.
In any of the above-described scheme preferably, the acquisition method of the offline labeled data is the following steps are included: step
01: since apart from crossing N meters, vehicle starts to acquire GPS data and image data in lane, wherein N is natural number；
Step 02: acquiring a frame image every Tms and record the differential GPS point coordinate data and vehicle at Image Acquisition moment
Road information, wherein T is natural number；
Step 03: from the road rightmost side, car lane is numbered to the left, lane number information where record current vehicle.
In any of the above-described scheme preferably, it acquires the offline labeled data and needs to meet following require:
1) the offline labeled data in each lane is acquired；
2) in collection process, vehicle driving need to lane center and travel speed be less than or equal to threshold speed.
In any of the above-described scheme preferably, the mask method of the offline labeled data the following steps are included:
Step 11: according to the collected data, every frame image being labeled, marks out image with rectangle frame in the picture
In area-of-interest；
Step 12: the GPS point coordinate information at the rectangle frame coordinate of mark acquisition moment corresponding with image is integrated into one 7
Tuple, [1., 2., 3., 4., 5., 6., 7.].
In any of the above-described scheme preferably, the interested region is rectangle frame, including four coordinate points difference
Upper left, lower-left, upper right and the lower-right most point coordinate of corresponding rectangle frame.
In any of the above-described scheme preferably, the rectangle frame need to cover the traffic lights region in image, and
Edge region has certain surplus space.
In any of the above-described scheme preferably, in 7 tuple, 1. and 2. for storing latitude and longitude coordinates point, 3.
For storing lane number information, it 4., 5., 6. and to be 7. used for storage rectangle frame coordinate.
In any of the above-described scheme preferably, the storage format of the offline labeled data is each crossing according to motor-driven
Lane quantity generates the storage file named with lane information, and every a line correspondence in the storage file stores described 7 yuan
Group data represent the area-of-interest position in corresponding lane in vehicle GPS coordinate downward view.
In any of the above-described scheme preferably, the application method of the offline labeled data is when intelligent vehicle is close to road
When mouth, according to the GPS information that vehicle positioning system positions, nearest point, acquisition pair are matched in acquired off-line data
The rectangle frame position for the area-of-interest answered.
In any of the above-described scheme preferably, the signal identification module is also used to training convolutional neural networks to traffic
The identification of signal lamp.
In any of the above-described scheme preferably, the color conversion formula in the traffic lights model are as follows:
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', g ', B '), Cmin=min (R ', G ',
B '), Δ=Cmax-Cmin, the normalized red component of R ' expression, R expression red component, the normalized green component of G ' expression, G
Indicate that green component, B ' indicate that normalized blue component, B indicate blue component, CmaxIndicate R ', G ', B after normalizing ' most
Big person, CminIndicate that the reckling of R ', G ', B after normalizing ', Δ indicate the difference of the maximum and reckling.
In any of the above-described scheme preferably, the signal identification module is also used to according to the traffic lights model
Extract red, green and yellow color region.
In any of the above-described scheme preferably, the recognition methods of the signal lamp the following steps are included:
Step 21: obtaining corresponding traffic signals region using GPS positioning；
Step 22: being detected in the traffic signals region using sliding window method, different sliding window sizes
It can obtain different rectangle frames；
Step 23: being screened using the rectangle frame of the non-maximum restraining method to overlapping
In any of the above-described scheme preferably, the non-maximum restraining method the following steps are included:
Step 231: in all bounding box outputs found out, rejecting the bounding box of pc≤0.6, pc is indicated ...；
Step 232: selecting the corresponding bounding box of maximum pc at this time, export the bounding box, and delete the bounding box；
Step 233: the corresponding bounding box of maximum pc at this time is selected again, and the calculating of loU is carried out with the bounding box of output,
If result >=0.5 of loU, not output boundary frame；Otherwise output boundary frame, and delete the bounding box；
Step 234: step 233 is repeated, until having handled all bounding boxes.
In any of the above-described scheme preferably, the camera is mounted on front windshield of vehicle middle position, and
And the camera position is fixed.
The invention proposes a kind of traffic lights recognition methods and system based on crossing priori knowledge, for specific
Crossing reduces the calculation amount of conventional mapping methods by the positioning of acquisition prior information additional transport signal lamp candidate region,
Trained depth model is recycled to carry out the identification of traffic lights in candidate region, accuracy with higher, simultaneously
This method is conducive to implement the identification of traffic lights in certain circumstances, especially in unmanned field.
LoU (Intersection over Union) refers to the intersection in union.
Detailed description of the invention
Fig. 1 is a preferred embodiment of the traffic lights recognition methods according to the invention based on crossing priori knowledge
Flow chart.
Fig. 2 is a preferred embodiment of the traffic lights identifying system according to the invention based on crossing priori knowledge
Module map.
Fig. 3 is another preferred embodiment of the traffic lights recognition methods according to the invention based on crossing priori knowledge
Process flow diagram.
Fig. 4 is the implementation as shown in Figure 3 of the traffic lights recognition methods according to the invention based on crossing priori knowledge
The area-of-interest figure of example.
Fig. 5 is the implementation as shown in Figure 3 of the traffic lights recognition methods according to the invention based on crossing priori knowledge
The non-maximum restraining method flow diagram of example.
Fig. 6 is the implementation as shown in Figure 3 of the traffic lights recognition methods according to the invention based on crossing priori knowledge
The recognition effect figure of example.
Specific embodiment
The present invention is further elaborated with specific embodiment with reference to the accompanying drawing.
Embodiment one
As shown in Figure 1, executing step 100, traffic scene image is inputted.
It executes step 110 and the traffic lights model of specific region is constructed according to the offline labeled data of acquisition.It is described
The acquisition method of offline labeled data is the following steps are included: step 01: since apart from crossing N meters, vehicle starts in lane
Acquire GPS data and image data, wherein N is natural number；Step 02: acquiring a frame image every Tms and record image and adopt
Collect the differential GPS point coordinate data and lane information at moment, wherein T is natural number；Step 03: from road rightmost side motor vehicle
Road is numbered to the left, lane number information where record current vehicle.Acquiring the offline labeled data needs satisfaction as follows
It is required that: 1) acquire the offline labeled data in each lane；2) in collection process, vehicle driving need to be in the center in lane
And travel speed is less than or equal to threshold speed.The mask method of the offline labeled data is the following steps are included: step 11: according to
The data of acquisition are labeled every frame image, mark out the area-of-interest in image with rectangle frame in the picture, described
Interested region is rectangle frame, and upper left, lower-left, upper right and the lower-right most point for respectively corresponding rectangle frame including four coordinate points are sat
Mark, the rectangle frame need to cover the traffic lights region in image, and edge region has certain surplus space；Step
12: the GPS point coordinate information at the rectangle frame coordinate of mark acquisition moment corresponding with image being integrated into 7 tuples, [1., 2.,
3. 4., 5., 6., 7.], in 7 tuple, 1. and 2. for storing latitude and longitude coordinates point, 3. for storing lane number letter
4., 5., 6. and 7. breath is used for storage rectangle frame coordinate.The storage format of the offline labeled data is each crossing according to machine
Motor-car road quantity generates the storage file named with lane information, and every a line correspondence in the storage file stores described 7
Tuple data represents the area-of-interest position in corresponding lane in vehicle GPS coordinate downward view.The offline labeled data
Application method be when intelligent vehicle is close to crossing, according to the GPS information that vehicle positioning system positions, it is acquired from
Line number matches nearest point in, obtains the rectangle frame position of corresponding area-of-interest.In the traffic lights model
Color conversion formula are as follows:
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', G ', B '), Cmin=min (R ', G ',
B '), Δ=Cmax-Cmin, the normalized red component of R ' expression, R expression red component, the normalized green component of G ' expression, G
Indicate that green component, B ' indicate that normalized blue component, B indicate blue component, CmaxIndicate R ', G ', B after normalizing ' most
Big person, CminIndicate that the reckling of R ', G ', B after normalizing ', Δ indicate the difference of the maximum and reckling.
Execute step 120, identification of the training convolutional neural networks to traffic lights, according to the traffic lights model
Extract red, green and yellow color region.The recognition methods of the signal lamp is the following steps are included: step 21: being obtained using GPS positioning
Corresponding traffic signals region；Step 22: being detected in the traffic signals region using sliding window method, different cunnings
Dynamic window size can obtain different rectangle frames；Step 23: being carried out using the rectangle frame of the non-maximum restraining method to overlapping
Screening.The non-maximum restraining method is the following steps are included: step 231: in all bounding box outputs found out, rejecting pc≤
0.6 bounding box, pc are indicated ...；Step 232: the corresponding bounding box of maximum pc at this time is selected, the bounding box is exported, and
Delete the bounding box；Step 233: selecting the corresponding bounding box of maximum pc at this time again, carry out loU's with the bounding box of output
It calculates, if result >=0.5 of loU, not output boundary frame；Otherwise output boundary frame, and delete the bounding box；Step 234:
Step 233 is repeated, until having handled all bounding boxes.
Step 130 is executed, recognition result is exported
Embodiment two
As shown in Fig. 2, a kind of traffic lights identifying system based on crossing priori knowledge, including camera 200, image
Input module 210, signal identification module 220 and output module.
Camera 200: being mounted on front windshield of vehicle middle position, while the camera position is fixed.
Image input module 210: for inputting traffic scene image.
Signal identification module 220: for the image according to input, traffic lights are identified.
Signal identification module 220 is used for the offline labeled data according to acquisition, constructs the traffic lights mould of specific region
Type.The acquisition method of the offline labeled data is the following steps are included: step 01: since apart from crossing N meters, vehicle is in lane
Inside start to acquire GPS data and image data, wherein N is natural number；Step 02: acquiring a frame image every Tms and record
The differential GPS point coordinate data and lane information at Image Acquisition moment, wherein T is natural number；Step 03: from the road rightmost side
Car lane is numbered to the left, lane number information where record current vehicle.The offline labeled data is acquired to need completely
Foot is following to be required: 1) acquiring the offline labeled data in each lane；2) in collection process, vehicle driving need to be in lane
Heart position and travel speed are less than or equal to threshold speed.The mask method of the offline labeled data is the following steps are included: step
11: according to the collected data, every frame image being labeled, marks out the region of interest in image with rectangle frame in the picture
Domain, interested region are rectangle frame, and upper left, lower-left, upper right and the lower-right most point of rectangle frame are respectively corresponded including four coordinate points
Coordinate, rectangle frame need to cover the traffic lights region in image, and edge region has certain surplus space；Step
12: the GPS point coordinate information at the rectangle frame coordinate of mark acquisition moment corresponding with image being integrated into 7 tuples, [1., 2.,
3. 4., 5., 6., 7.], in 7 tuple, 1. and 2. for storing latitude and longitude coordinates point, 3. for storing lane number letter
4., 5., 6. and 7. breath is used for storage rectangle frame coordinate.The storage format of the offline labeled data is each crossing according to machine
Motor-car road quantity generates the storage file named with lane information, and every a line correspondence in the storage file stores described 7
Tuple data represents the area-of-interest position in corresponding lane in vehicle GPS coordinate downward view.Offline labeled data is answered
It is when intelligent vehicle is close to crossing, according to the GPS information that vehicle positioning system positions, in acquired offline number with method
According to the nearest point of middle matching, the rectangle frame position of corresponding area-of-interest is obtained.
Signal identification module 220 is also used to identification of the training convolutional neural networks to traffic lights.The traffic signals
Color conversion formula in lamp model are as follows:
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', G ', B '), Cmin=min (R ', G ',
B '), Δ=Cmax-Cmin, the normalized red component of R ' expression, R expression red component, the normalized green component of G ' expression, G
Indicate that green component, B ' indicate that normalized blue component, B indicate blue component, CmaxIndicate R ', G ', B after normalizing ' most
Big person, CminIndicate that the reckling of R ', G ', B after normalizing ', Δ indicate the difference of the maximum and reckling.
Signal identification module 220 is also used to red, the green and yellow color region according to the traffic lights model extraction.It is described
The recognition methods of signal lamp is the following steps are included: step 21: obtaining corresponding traffic signals region using GPS positioning；Step 22:
It is detected in the traffic signals region using sliding window method, different sliding window sizes can obtain different rectangles
Frame；Step 23: being screened using the rectangle frame of the non-maximum restraining method to overlapping.The non-maximum restraining method includes
Following steps: step 231: in all bounding box outputs found out, rejecting the bounding box of pc≤0.6, pc is indicated ...；Step
Rapid 232: selecting the corresponding bounding box of maximum pc at this time, export the bounding box, and delete the bounding box；Step 233: selecting again
The corresponding bounding box of maximum pc at this time out carries out the calculating of loU with the bounding box of output, if result >=0.5 of loU,
Not output boundary frame；Otherwise output boundary frame, and delete the bounding box；Step 234: step 233 is repeated, it is all until having handled
Bounding box.
Output module 230: output recognition result.
Embodiment three
As shown in figure 3, the process flow of the traffic lights recognition methods based on crossing priori knowledge is as follows:
Preparation stage: vehicle-mounted camera is mounted on front windshield of vehicle middle position, and necessary camera position is solid
It is fixed.
Method and step:
Off-line phase:
It acquires in crossing cartographic information (region GPS+ROI), acquisition method+data record format
Acquisition method: since apart from 100 meters of crossing, vehicle starts to acquire GPS data and image data in lane；Often
A frame image is acquired every 100ms and records the differential GPS point coordinate data and lane information at Image Acquisition moment, most from road
Right side car lane is numbered to the left, lane number information where record current vehicle.It will be repeated for every lane
The acquisition step in face, it is desirable that vehicle driving is in the center in lane, travel speed≤35km/h.
Mask method: according to the collected data, every frame image is labeled, marks out figure with rectangle frame in the picture
Area-of-interest as in (four coordinate points are upper left and the lower-right most point coordinate of rectangle frame respectively), it is desirable that the rectangle frame of mark
The traffic lights region in image can be covered, and edge region has certain surplus space, by the rectangle frame of mark
The GPS point coordinate information at coordinate acquisition moment corresponding with image is integrated into 7 tuples, [1., 2., 3., 4., 5., 6., 7.],
2. 4. 5. 6. 7. storage rectangle frame is sat 1. storage latitude and longitude coordinates point (after decimal point eight), 3. stores lane number information
Mark.
Offline labeled data storage format: each crossing generates the txt named with lane information according to car lane quantity
File, corresponding 7 tuple datas storing above-mentioned mark and generating of every a line in txt file, represents the vehicle in corresponding lane
Area-of-interest position in GPS coordinate downward view.
The application of off-line data: when intelligent vehicle is close to crossing, according to the GPS information that vehicle positioning system positions,
Nearest point (range of threshold value, 10cm, 0.000001) is matched in acquired off-line data, obtains corresponding region of interest
The rectangle frame position (as shown in Figure 4) in domain, then further realizes the detection and identification to traffic lights in the rectangle frame.
The cognitive phase of ventilating signal lamp:
1, the traffic lights model of specific region is constructed, according to the collected data, for detecting the traffic lights in scene
Region further reduces area-of-interest.(expansion correlative detail)
RGB- > hsv color space, green, red color model ()
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', G ', B '), Cmin=min (R ', G ',
B '), Δ=Cmax-Cmin。
It is as follows to extract parameter used in red, green, yellow color region process:
Red color range: [0,100,100]~[10,255,255]；[160,100,100]~[180,255,255]；
Green fields: [40,50,50]~[90,255,255]；
Yellow: [15,150,150]~[35,255,255]
2, on the basis of previous step, training convolutional neural networks (GoogleLenet) roll up the identification of traffic lights
The details of product neural metwork training, parameter.Including data set, quantity, (expansion correlative detail)
Average_loss:40
Base_lr:0.01
Power:0.5
Momentum:0.9
Weight_decay:0.0002
During traversing traffic signals to the regional area in traffic scene using the model of convolutional neural networks training,
The setting of sliding window:
Sliding window size: [(20,46), (30,69), (40,92), (50,115), (60,138)], it can also be according to reality
Border image size adjust sliding window size, it then follows rule be transverse and longitudinal ratio: 1: 2.2~1: 2.5, IoU's is dimensioned to
0.7, testing result probability 0.7, step-length: 8 pixels.
Non-maximum restraining (non-max suppression), exactly in order to propose duplicate object detection results, step is such as
Shown in Fig. 5:
1) in all bounding box outputs found out, the bounding box of pc≤0.6 is rejected.Due to smaller than 0.6, expression is complete here
Object out is that the probability of target object is little, therefore does not consider.
2) the corresponding bounding box of maximum pc at this time is selected, the bounding box is exported, and deletes the bounding box.
3) the corresponding bounding box of maximum pc at this time is selected again, and the calculating of loU is carried out with the bounding box of output, if
Result >=0.5 of loU, then it is assumed that this bounding box is similar with the bounding box exported, then not output boundary frame；Otherwise it exports
Bounding box.After being disposed, the bounding box is rejected.
4) step 3 is repeated, until having handled all bounding boxes.
Algorithm recognition effect is as shown in Figure 6.
For a better understanding of the present invention, the above combination specific embodiments of the present invention are described in detail, but are not
Limitation of the present invention.Any simple modification made to the above embodiment according to the technical essence of the invention, still belongs to
In the range of technical solution of the present invention.In this specification the highlights of each of the examples are it is different from other embodiments it
Locate, the same or similar part cross-reference between each embodiment.For system embodiments, due to itself and method
Embodiment corresponds to substantially, so being described relatively simple, the relevent part can refer to the partial explaination of embodiments of method.
Claims (10)
1. a kind of traffic lights recognition methods based on crossing priori knowledge, including input traffic scene image, feature exist
In further comprising the steps of:
Step 1: according to the offline labeled data of acquisition, constructing the traffic lights model of specific region；
Step 2: identification of the training convolutional neural networks to traffic lights；
Step 3: output recognition result.
2. as described in claim 1 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that it is described from
The acquisition method of line labeled data the following steps are included:
Step 01: since apart from crossing N meters, vehicle starts to acquire GPS data and image data in lane, wherein N be from
So number；
Step 02: acquiring a frame image every Tms and record the differential GPS point coordinate data and lane letter at Image Acquisition moment
Breath, wherein T is natural number；
Step 03: from the road rightmost side, car lane is numbered to the left, lane number information where record current vehicle.
3. as claimed in claim 2 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that acquisition institute
It states offline labeled data and needs to meet following require:
1) the offline labeled data in each lane is acquired；
2) in collection process, vehicle driving need to lane center and travel speed be less than or equal to threshold speed.
4. as claimed in claim 3 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that it is described from
The mask method of line labeled data the following steps are included:
Step 11: according to the collected data, every frame image being labeled, is marked out in image with rectangle frame in the picture
Area-of-interest；
Step 12: the GPS point coordinate information at the rectangle frame coordinate of mark acquisition moment corresponding with image is integrated into one 7 yuan
Group, [1., 2., 3., 4., 5., 6., 7.].
5. as claimed in claim 4 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that the sense
The region of interest is rectangle frame, and upper left, lower-left, upper right and the lower-right most point coordinate of rectangle frame are respectively corresponded including four coordinate points.
6. as claimed in claim 5 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that the square
Shape frame need to cover the traffic lights region in image, and edge region has certain surplus space.
7. as claimed in claim 4 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that described
In 7 tuples, 1. and 2. for storing latitude and longitude coordinates point, 3. for storing lane number information, 4., 5., 6. and 7. for depositing
Store up rectangle frame coordinate.
8. as claimed in claim 7 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that it is described from
The storage format of line labeled data is each crossing according to car lane quantity, generates the storage file named with lane information,
Every a line in the storage file, which corresponds to, stores 7 tuple data, represents and regards under vehicle GPS coordinate in corresponding lane
The area-of-interest position of Yezhong.
9. as described in claim 1 based on the traffic lights recognition methods of crossing priori knowledge, which is characterized in that the friendship
Color conversion formula in ventilating signal lamp model are as follows:
V=Cmax
Wherein: R '=R/255, G '=G/255, B '=B/255, Cmax=max (R ', G ', B '), Cmin=min (R ', G ', B '),
Δ=Cmax-Cmin, the normalized red component of R ' expression, R expression red component, the normalized green component of G ' expression, G expression
Green component, B ' indicate that normalized blue component, B indicate blue component, CmaxIndicate the maximum of R ', G ', B after normalizing '
Person, CminIndicate that the reckling of R ', G ', B after normalizing ', Δ indicate the difference of the maximum and reckling.
10. a kind of traffic lights identifying system based on crossing priori knowledge, including camera and image input module, special
Sign is, further includes with lower module:
Signal identification module: for the image according to input, traffic lights are identified.
Output module: output recognition result；
The system is implemented according to the method for claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910189015.XA CN109949593A (en) | 2019-03-13 | 2019-03-13 | A kind of traffic lights recognition methods and system based on crossing priori knowledge |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910189015.XA CN109949593A (en) | 2019-03-13 | 2019-03-13 | A kind of traffic lights recognition methods and system based on crossing priori knowledge |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109949593A true CN109949593A (en) | 2019-06-28 |
Family
ID=67009679
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910189015.XA Pending CN109949593A (en) | 2019-03-13 | 2019-03-13 | A kind of traffic lights recognition methods and system based on crossing priori knowledge |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109949593A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287244A (en) * | 2019-07-03 | 2019-09-27 | 武汉中海庭数据技术有限公司 | It is a kind of based on the traffic lights localization method repeatedly clustered |
| CN110688992A (en) * | 2019-12-09 | 2020-01-14 | 中智行科技有限公司 | Traffic signal identification method and device, vehicle navigation equipment and unmanned vehicle |
| CN110728170A (en) * | 2019-08-08 | 2020-01-24 | 北京联合大学 | Hybrid model traffic signal detection method and system based on intersection information |
| CN112133088A (en) * | 2020-08-25 | 2020-12-25 | 浙江零跑科技有限公司 | Vehicle traffic auxiliary indication method and system |
| CN112751633A (en) * | 2020-10-26 | 2021-05-04 | 中国人民解放军63891部队 | Broadband spectrum detection method based on multi-scale window sliding |
| CN112880692A (en) * | 2019-11-29 | 2021-06-01 | 北京市商汤科技开发有限公司 | Map data annotation method and device and storage medium |
| CN115249407A (en) * | 2021-05-27 | 2022-10-28 | 上海仙途智能科技有限公司 | Indicating lamp state identification method and device, electronic equipment, storage medium and product |
| CN115394103A (en) * | 2022-07-29 | 2022-11-25 | 阿波罗智联(北京)科技有限公司 | Method, device, equipment and storage medium for identifying signal lamp |
| US11790772B2 (en) | 2020-09-23 | 2023-10-17 | Apollo Intelligent Connectivity (Beijing) Technology Co., Ltd | Traffic light image processing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105489035A (en) * | 2015-12-29 | 2016-04-13 | 大连楼兰科技股份有限公司 | Detection method of traffic lights applied to active drive technology |
| CN105930819A (en) * | 2016-05-06 | 2016-09-07 | 西安交通大学 | System for real-time identifying urban traffic lights based on single eye vision and GPS integrated navigation system |
| CN106023605A (en) * | 2016-07-15 | 2016-10-12 | 姹ゅ钩 | Traffic signal lamp control method based on deep convolution neural network |
| CN107038420A (en) * | 2017-04-14 | 2017-08-11 | 北京航空航天大学 | A kind of traffic lights recognizer based on convolutional network |
-
2019
- 2019-03-13 CN CN201910189015.XA patent/CN109949593A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105489035A (en) * | 2015-12-29 | 2016-04-13 | 大连楼兰科技股份有限公司 | Detection method of traffic lights applied to active drive technology |
| CN105930819A (en) * | 2016-05-06 | 2016-09-07 | 西安交通大学 | System for real-time identifying urban traffic lights based on single eye vision and GPS integrated navigation system |
| CN106023605A (en) * | 2016-07-15 | 2016-10-12 | 姹ゅ钩 | Traffic signal lamp control method based on deep convolution neural network |
| CN107038420A (en) * | 2017-04-14 | 2017-08-11 | 北京航空航天大学 | A kind of traffic lights recognizer based on convolutional network |
Non-Patent Citations (1)
| Title |
|---|
| 王坚: "基于深度属性学习的交通标志识别方法研究", 《中国优秀硕士学位论文全文数据库（信息科技辑）》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110287244B (en) * | 2019-07-03 | 2021-03-16 | 武汉中海庭数据技术有限公司 | Traffic light positioning method based on multiple clustering |
| CN110287244A (en) * | 2019-07-03 | 2019-09-27 | 武汉中海庭数据技术有限公司 | It is a kind of based on the traffic lights localization method repeatedly clustered |
| CN110728170B (en) * | 2019-08-08 | 2023-08-18 | 北京联合大学 | Intersection information-based traffic signal detection method and system of mixed model |
| CN110728170A (en) * | 2019-08-08 | 2020-01-24 | 北京联合大学 | Hybrid model traffic signal detection method and system based on intersection information |
| CN112880692A (en) * | 2019-11-29 | 2021-06-01 | 北京市商汤科技开发有限公司 | Map data annotation method and device and storage medium |
| CN112880692B (en) * | 2019-11-29 | 2024-03-22 | 北京市商汤科技开发有限公司 | Map data labeling method and device and storage medium |
| CN110688992A (en) * | 2019-12-09 | 2020-01-14 | 中智行科技有限公司 | Traffic signal identification method and device, vehicle navigation equipment and unmanned vehicle |
| CN112133088A (en) * | 2020-08-25 | 2020-12-25 | 浙江零跑科技有限公司 | Vehicle traffic auxiliary indication method and system |
| US11790772B2 (en) | 2020-09-23 | 2023-10-17 | Apollo Intelligent Connectivity (Beijing) Technology Co., Ltd | Traffic light image processing |
| CN112751633A (en) * | 2020-10-26 | 2021-05-04 | 中国人民解放军63891部队 | Broadband spectrum detection method based on multi-scale window sliding |
| CN115249407A (en) * | 2021-05-27 | 2022-10-28 | 上海仙途智能科技有限公司 | Indicating lamp state identification method and device, electronic equipment, storage medium and product |
| CN115249407B (en) * | 2021-05-27 | 2023-09-26 | 上海仙途智能科技有限公司 | Indicator light state identification method and device, electronic equipment, storage medium and product |
| CN115394103A (en) * | 2022-07-29 | 2022-11-25 | 阿波罗智联(北京)科技有限公司 | Method, device, equipment and storage medium for identifying signal lamp |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109949593A (en) | A kind of traffic lights recognition methods and system based on crossing priori knowledge | |
| CN105930819B (en) | Real-time city traffic lamp identifying system based on monocular vision and GPS integrated navigation system | |
| Kong et al. | General road detection from a single image | |
| Rasmussen | Combining laser range, color, and texture cues for autonomous road following | |
| CN104134234B (en) | A kind of full automatic three-dimensional scene construction method based on single image | |
| CN109949316A (en) | A kind of Weakly supervised example dividing method of grid equipment image based on RGB-T fusion | |
| CN109284669A (en) | Pedestrian detection method based on Mask RCNN | |
| CN102999753B (en) | License plate locating method | |
| CN106326858A (en) | Road traffic sign automatic identification and management system based on deep learning | |
| CN105261017A (en) | Method for extracting regions of interest of pedestrian by using image segmentation method on the basis of road restriction | |
| CN108805018A (en) | Road signs detection recognition method, electronic equipment, storage medium and system | |
| CN103544505B (en) | Ship identification system and method towards unmanned plane Aerial Images | |
| CN111915583B (en) | Vehicle and pedestrian detection method based on vehicle-mounted thermal infrared imager in complex scene | |
| CN107491762A (en) | A kind of pedestrian detection method | |
| CN104102909B (en) | Vehicle characteristics positioning and matching process based on lenticular information | |
| CN104463138B (en) | The text positioning method and system of view-based access control model structure attribute | |
| CN106874884A (en) | Human body recognition methods again based on position segmentation | |
| Wang et al. | An overview of 3d object detection | |
| CN112381870B (en) | Binocular vision-based ship identification and navigational speed measurement system and method | |
| CN107194343B (en) | Traffic lights detection method based on the relevant convolution in position Yu Fire model | |
| CN111414954B (en) | Rock image retrieval method and system | |
| CN107704853A (en) | A kind of recognition methods of the traffic lights based on multi-categorizer | |
| CN107944403A (en) | Pedestrian's attribute detection method and device in a kind of image | |
| CN107369158A (en) | The estimation of indoor scene layout and target area extracting method based on RGB D images | |
| CN109961013A (en) | Recognition methods, device, equipment and the computer readable storage medium of lane line |
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: 20190628 |