CN104091344B - Road dividing method - Google Patents
Road dividing method Download PDFInfo
- Publication number
- CN104091344B CN104091344B CN201410350481.9A CN201410350481A CN104091344B CN 104091344 B CN104091344 B CN 104091344B CN 201410350481 A CN201410350481 A CN 201410350481A CN 104091344 B CN104091344 B CN 104091344B
- Authority
- CN
- China
- Prior art keywords
- training
- road
- mlp
- road image
- gab
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013507 mapping Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 38
- 230000011218 segmentation Effects 0.000 claims description 23
- 241000209094 Oryza Species 0.000 claims description 8
- 235000007164 Oryza sativa Nutrition 0.000 claims description 8
- 235000009566 rice Nutrition 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 7
- 230000008447 perception Effects 0.000 claims description 6
- 238000001514 detection method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 239000013598 vector Substances 0.000 description 5
- 238000011161 development Methods 0.000 description 4
- 230000018109 developmental process Effects 0.000 description 4
- 230000007613 environmental effect Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 206010039203 Road traffic accident Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000004134 energy conservation Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000004642 transportation engineering Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a road dividing method. The method comprises the steps that S1, a multilayer perceptron is trained offline to fit the mapping relation between the characteristics of adjacent pixel points and an n-link weight so as to estimate the n-link weight, and a gentle adaboost classifier is trained offline to fit the mapping relation between the area characteristics of the pixel points and a t-link weight so as to estimate the t-link weight; and S2, the n-link weight and the t-link weight are estimated through the well-trained multilayer perceptron and the well-trained gentle adaboost classifier respectively, a road image is divided online through an s-t image division method, and a road area in the road image is obtained. The road dividing method has good robustness for interference factors in the road environment.
Description
Technical field
The present invention relates to vehicle assistant drive technical field, particularly a kind of lane segmentation side for road detection system
Method.
Background technology
The development of auto industry has had the history of more than 100 years, the appearance of automobile greatly improve the production of people and
Life style, promotes the development of social economy.Automobile also brings problems while offering convenience for human lives,
Such as traffic accident takes place frequently, and energy resource consumption is too fast, and environmental pollution is serious etc..In the urban district of traffic congestion, driver's average minute
Clock needs to complete manipulation of the 20-30 hand-foot coordination action to realize vehicle, therefore driving task is phase under traffic congestion
When complicated.With the development of society, requirement of the people to automobile in security, the feature of environmental protection, economy and comfortableness is also got over
Come higher.Sensor, computer, automatically control, the technology such as artificial intelligence and machine vision in constantly development and is reformed, and
More and more apply in Communication and Transportation Engineering.This background has expedited the emergence of the research and development of intelligent vehicle, vehicle is possessed independently
Drive and aid in the function of driving, to realize safety, energy-conservation, convenient, comfortable driving experience.
Road detection system is the important component part of intelligent vehicle DAS (Driver Assistant System).Intelligent vehicle passes through Road Detection
System can traffic areas and the vehicle body position of road boundary and attitude relatively to obtain.So, can be realized by Road Detection
Auxiliary automobile navigation, lane departure warning, track keeps, and adaptive cruise monitors the driving condition of driver, predicts driver
Behavior be intended to etc. it is many auxiliary drive function.
At present most Approach for road detection is based on computer vision.In these methods, by road area picture
The process that vegetarian refreshments separates with other area pixel points is referred to as lane segmentation.Lane segmentation is a challenging problem：
On the one hand, due to the impact of the factors such as road surface material, weather condition, illumination variation, road surface has various outward appearances；It is another
Aspect, with the motion of vehicle, road surface and background are, in dynamic change, and vehicle and pedestrian etc. to be generally there are on road surface
Disturbing factor.Above-mentioned factor is easily impacted to the accuracy of lane segmentation, is that lane segmentation brings extreme difficulties.
The content of the invention
(1) technical problem is solved
The invention aims to the technical problem that current lane segmentation method is easily subject to environmental factor interference is solved,
Therefore, the present invention proposes a kind of lane segmentation method of robust.
(2) technical scheme
It is as follows to solve above-mentioned technical problem, a kind of lane segmentation method of present invention proposition, including step：
Step S1：Off-line training multi-layer perception (MLP) (Multilayer Perceptron, MLP) is being fitted neighbor pixel
Feature to n-link weights mapping relations, to estimate n-link weights, off-line training Gentle Adaboost (GAB)
Grader, carrys out match pixel neighborhood of a point feature to the mapping relations of t-link weights, to estimate t-link weights；
Step S2：N-link weights and t-link weights are estimated respectively using the MLP and GAB that train, using s-t figures
Figure segmentation method is split online to road image, obtains the road area in road image.
(3) beneficial effect
The present invention carries out lane segmentation under the method frame that figure cuts, and has closely combined the global information in road image
And local message, there is preferable robustness to local interference；The present invention is fitted the feature and n- of neighbor pixel by MLP
Mapping relations between link weights, can make up the contrast meter of neighbor pixel used in traditional method cut based on figure
The mode of n-link weights is calculated easily by the shortcoming of environmental factor interference；The present invention is by GAB come match pixel neighborhood of a point feature
To the mapping relations of t-link weights, can weaken that road surface outward appearance is changeable and background dynamics variation characteristic is for the shadow of lane segmentation
Ring, improve the accuracy of segmentation.Therefore, the present invention has preferable robustness to the disturbing factor in road environment.
Description of the drawings
Fig. 1 is the flow chart of lane segmentation of the present invention；
Fig. 2 is the connection diagram of adjacent node in s-t figures of the present invention；
Fig. 3 is s-t figures connection diagram of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention become more apparent, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in more detail.
The figure that the present invention proposes a kind of n-link weights for lane segmentation and can learn cuts method, by road image s-
T figures (s-t Graph) expressions, realize Road image segmentation, as shown in figure 1, institute in the case where figure cuts the method frame of (Graph Cut)
The method of stating includes：
Step S1：Off-line training MLP being fitted the mapping relations of the feature of neighbor pixel to n-link weights, to
Estimate n-link weights, off-line training GAB carrys out match pixel neighborhood of a point feature to the mapping relations of t-link weights, to
Estimate t-link (Terminal Link) weights.
Step S2：N-link weights and t-link weights are estimated respectively using the MLP and GAB that train, using s-t figures
Figure segmentation method is split online to road image, obtains the road area in road image.
It is a kind of combined optimization method based on graph theory that described figure cuts, and has in computer vision field and widely should
With.When figure cuts is applied to image segmentation, image to be split is levied with s-t charts, with max-flow/minimal cut (Max-flow/
Min-cut) algorithm asks for the minimal cut of s-t figures.The process for asking for minimal cut is equal to the process for minimizing an energy function,
Obtain minimal cut and optimum segmentation of the image under the energy function is obtained.
The training process of the MLP is comprised the following steps：
Step S1A：Training sample is extracted from the road image sample set for marking；
Because pixel is numerous in road image, it is not necessary that all pixels point in the image that extraction has been marked is used as instruction
Practice sample.On the one hand, extracting excessive sample can bring the sharp increase of amount of calculation；On the other hand, excessive sample is difficult to can MLP
Focus on some disturbing factors, such as shade, hot spot, lane markings etc..It is therefore desirable to using a kind of sampling policy from mark
Sample is extracted in good image.Only adjacent from the region of the road surface pixel of positive sample, negative sample extracts from the boundary of road.
It is not because that we are indifferent to the n-link weights in non-rice habitats region from non-rice habitats extracted region positive sample, and reduces data
Diversity is conducive to the convergence of MLP.High-contrast area, high brightness in the present invention according to certain probability from road area
Region and general area randomly select positive sample, and the probability of institute's foundation determines ratio of each area sample in positive sample, enters
And determine powers of the MLP for various different road conditions adaptability.Pixel on road boundary is all by as negative sample
This, because the number of these pixels is far smaller than the number of road area pixel.The corresponding estimator output power of positive sample
It is worth for 1, the corresponding estimator output weights of negative sample are 0.Each training sample for being extracted is the spy of one 10 × 1 dimension
Vector is levied, included feature and its definition is as shown in table 1.
The input feature vector of the multi-layer perception (MLP) of table 1
Step S1B：Using the training sample training MLP for extracting, until its training error is in receptible scope, obtain
The parameter of MLP.
Mapping relations between the feature and n-link weights of neighbor pixel are learnt using MLP, for estimating n-link
Weights.Road image is represented the present invention adjacent node in s-t figures is connected by way of eight connectivity with s-t figures, is such as schemed
In 2 shown in (a), adjacent node by level, vertical be connected with diagonal side.Also, (a) is segmented into multiple by one in Fig. 2
The elementary cell of individual node and four edges composition, in such as Fig. 2 shown in (b).Certainly, the node in image border does not have described
Four edges, but can lack while regard as weights be 0 while.So, can with four independent MLP to be fitted Fig. 2 in
The mapping relations of the weights of the n-link of the four direction shown in (b).Each MLP is one four layers of BP Neural Network
Network, including an input layer, two hidden layers and an output layer, respectively comprising 10,20,20,1 neurons.Each neuron
Activation primitive be sigmoid functions.The characteristic vector of one 10 × 1 dimension of input of each MLP, its definition is as shown in table 1.
MLP output areas are that between 0 to 1, output is closer to 1, and corresponding n-link weights are bigger.
By RPROP algorithms, using the training sample training multi-layer perception (MLP) extracted in step S1A.In order to eliminate positive sample
Impact of the difference of this and negative sample quantity to training result, balance hits rate and false alarm rate, needs to be negative sample in training
Addition weight, weight coefficient is as follows：
Wherein, w (y) be for negative sample addition weight coefficient, NpAnd NnThe number of positive sample and negative sample, y are represented respectively
It is the class label of sample, y=1 is positive sample, is otherwise negative sample.
The training process of the GAB is comprised the following steps：
Step S1a：Training sample is extracted from the road image sample set for marking；
According to certain ratio from road area and non-rice habitats extracted region positive sample and negative sample.Each sample is one
The characteristic vectors of individual 13 × 1 dimension, comprising feature have R, G, B value of pixel, in 9 × 9 windows centered on the pixel
The average and variance of R, G, B value, the Grad and gradient direction of the pixel, and its coordinate under image coordinate system.Positive sample
This corresponding output valve is 1, and the corresponding output valve of negative sample is -1.
Step S1b：Using the training sample training GAB for extracting, until its training error is in receptible scope, obtain
The parameter of t-link weights estimation devices.
Using the mapping relations of GAB graders match pixel neighborhood of a point feature to t-link weights.Because GAB is
One mutation algorithm of Adaboost algorithm, with good feature selecting ability and numerical stability, for noise has more preferably
Robustness, and GAB is a kind of good homing method for having a supervision.The GAB graders are made up of 500 decision-making stubs,
Each decision-making stub can be regarded as the decision tree of an only one of which node.The output of GAB is not class label, but all
The weighted sum of the votes of Weak Classifier.For positive sample, it is intended that GAB output on the occasion of；For negative sample, it is intended that
GAB exports negative value.Using the training sample training GAB graders extracted in step Sla.
The online cutting procedure of road image is comprised the following steps：
Step S21：Road image I is obtained by vehicle-mounted imageing sensor；
Step S22：S-t figures are built, each pixel in road image I is set to into adjacent node (Neighborhood
Nodes), adjacent node is connected by way of four connections or eight connectivity with adjacent node, source node (Source Node) generation
Table road area is simultaneously set to prospect, and sink nodes (Sink Node) represent non-rice habitats region and are set to background, each adjacent node point
It is not connected with source node and sink nodes；S-t figures are a kind of graph models, including two kinds of nodes and two kinds of sides.Two kinds of nodes are respectively
Adjacent node and end node, end node includes source node and convergence point.
Connection between s-t figure interior joints as shown in figure 3, s signs for source node, t signs for sink nodes, it is black
Color solid dot is adjacent node.The eight connectivity mode that adjacent node passes through is connected, and the weights in each edge represent the side two ends top
The correlation degree of point, the weights on the side that adjacent node is connected with source node represent adjacent node as the general of the pixel of road area
Rate, the weights on the side that adjacent node is connected with sink nodes represent adjacent node as the probability of non-rice habitats area pixel point.
Step S23：Using the MLP for training estimate respectively in road image I each adjacent node be adjacent node it
Between connection side weights；
Each adjacent node needs the side for estimating weights to have its left side adjacent node, and upper left side adjacent node, top is adjacent
Node and upper right side adjacent node connection side, in such as Fig. 2 shown in (b).Node in image border does not have described four edges,
But can lack while regard as weights be 0 while.The feature (definition is as shown in table 1) of each adjacent node is extracted, by spy
Levy and be input in corresponding MLP.The output of MLP is the weights of corresponding sides, i.e.,：
B(p, q)=MLPd(vP, q), d=dir (p, q) (2)
Wherein, B(p, q)It is multi-layer perception (MLP) MLPd(vP, q) output corresponding sides weights, MLPd(vP, q) represent for estimating
Count the multi-layer perception (MLP) of the n-link weights on d-th direction, vP, qIt is the characteristic vector of input, dir (p, q) represents adjacent picture
Plain p and q connects the direction on side.
Step S24：Estimate that each adjacent node is connected side with source node in road image I respectively using the GAB for training
Weights, each adjacent node be connected the weights on side with sink nodes；
The feature of each adjacent node is extracted, the GAB graders for training are input into.GAB graders export all weak typings
The weighted sum of the votes of device, in addition it is also necessary to which, further across change of scale to interval [0,1], the weight calculation for connecting side is as follows：
Wherein, lpFor the class label of pixel p, w is normal number, xpIt is outputs of the GAB at pixel p, lpR when=1p
(xp, lp) for pixel p to sink nodes connect side weight, lpR when=0p(xp, lp) it is connections of the pixel p to source node
The weight on side.
Step S25：The minimal cut of the s-t figures for asking for building using max-flow/minimal cut algorithm, in obtaining road image I
The class label L={ l of each pixeli|li∈ { 0,1 }, i=1,2 ... N }, wherein, N is the individual of pixel in road image I
Number, liFor the class label of pixel i, li=1 represents pixel i as road area pixel, and no person is non-rice habitats area pixel
Point.
Particular embodiments described above, has been carried out further in detail to the purpose of the present invention, technical scheme and beneficial effect
Describe in detail it is bright, should be understood that the foregoing is only the present invention specific embodiment, be not limited to the present invention, it is all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvements done etc., should be included in the guarantor of the present invention
Within the scope of shield.
Claims (6)
1. a kind of lane segmentation method, including step is as follows：
Step S1：Off-line training multi-layer perception (MLP) MLP, to estimate n-link weights, off-line training Gentle Adaboost divide
Class device GAB, to estimate t-link weights；
Step S2：N-link weights and t-link weights are estimated respectively using the MLP and GAB that train, are cut using the figure of s-t figures
Method is split online to road image, obtains the road area in road image.
2. lane segmentation method as claimed in claim 1, it is characterised in that the effect of the MLP is fitting neighbor pixel
Feature to n-link weights mapping relations.
3. lane segmentation method as claimed in claim 1, it is characterised in that the process of the training MLP is comprised the following steps：
Step S1A：Training sample is extracted from the road image sample set for marking；
Step S1B：Using the training sample training MLP for extracting, until its training error is in receptible scope, MLP is obtained
Parameter.
4. lane segmentation method as claimed in claim 1, it is characterised in that the effect of the GAB is the neighbour of match pixel point
Mapping relations of the characteristic of field to t-link weights.
5. lane segmentation method as claimed in claim 1, it is characterised in that the process of the training GAB is comprised the following steps：
Step S1a：Training sample is extracted from the road image sample set for marking；
Step S1b：Using the training sample training GAB for extracting, until its training error is in receptible scope, GAB is obtained
Parameter.
6. lane segmentation method as claimed in claim 1, it is characterised in that road image the is split mistake online
Journey is comprised the following steps：
Step S21：Road image is obtained by vehicle-mounted imageing sensor；
Step S22：Build s-t figures, each pixel in road image be set to into adjacent node, adjacent node by four connections or
The mode of person's eight connectivity is connected with adjacent node, and source node represents road area and is set to prospect, and sink nodes represent non-rice habitats area
Domain is simultaneously set to background, and each adjacent node is connected respectively with source node and sink nodes；
Step S23：Estimate that each adjacent node is adjacent between node in road image I respectively using the MLP for training
The weights on connection side；
Step S24：Estimate that each adjacent node in road image is connected the power on side with source node respectively using the GAB for training
Value, each adjacent node are connected the weights on side with sink nodes；
Step S25：The minimal cut of the s-t figures for asking for building using max-flow/minimal cut algorithm, obtains each picture in road image
The class label L={ l of vegetarian refreshmentsi|li∈ { 0,1 }, i=1,2 ... N }, wherein, N be road image in pixel number, liIt is picture
The class label of vegetarian refreshments i, li=1 represents pixel i as road area pixel, is otherwise non-rice habitats area pixel point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410350481.9A CN104091344B (en) | 2014-07-22 | 2014-07-22 | Road dividing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410350481.9A CN104091344B (en) | 2014-07-22 | 2014-07-22 | Road dividing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104091344A CN104091344A (en) | 2014-10-08 |
| CN104091344B true CN104091344B (en) | 2017-04-19 |
Family
ID=51639059
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410350481.9A Active CN104091344B (en) | 2014-07-22 | 2014-07-22 | Road dividing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104091344B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105631880B (en) * | 2015-12-31 | 2019-03-22 | 百度在线网络技术（北京）有限公司 | Lane line dividing method and device |
| CN106558058B (en) * | 2016-11-29 | 2020-10-09 | 北京图森未来科技有限公司 | Segmentation model training method, road segmentation method, vehicle control method and device |
| CN108229274B (en) * | 2017-02-28 | 2020-09-04 | 北京市商汤科技开发有限公司 | Method and device for training multilayer neural network model and recognizing road characteristics |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101558404A (en) * | 2005-06-17 | 2009-10-14 | 微软公司 | Image segmentation |
| CN102609686A (en) * | 2012-01-19 | 2012-07-25 | 宁波大学 | Pedestrian detection method |
| CN103473767A (en) * | 2013-09-05 | 2013-12-25 | 中国科学院深圳先进技术研究院 | Segmentation method and system for abdomen soft tissue nuclear magnetism image |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7212201B1 (en) * | 1999-09-23 | 2007-05-01 | New York University | Method and apparatus for segmenting an image in order to locate a part thereof |
| US6961454B2 (en) * | 2001-10-04 | 2005-11-01 | Siemens Corporation Research, Inc. | System and method for segmenting the left ventricle in a cardiac MR image |
| US7400757B2 (en) * | 2001-10-04 | 2008-07-15 | Siemens Medical Solutions Usa, Inc. | System and method for segmenting the left ventricle in a cardiac image |
-
2014
- 2014-07-22 CN CN201410350481.9A patent/CN104091344B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101558404A (en) * | 2005-06-17 | 2009-10-14 | 微软公司 | Image segmentation |
| CN102609686A (en) * | 2012-01-19 | 2012-07-25 | 宁波大学 | Pedestrian detection method |
| CN103473767A (en) * | 2013-09-05 | 2013-12-25 | 中国科学院深圳先进技术研究院 | Segmentation method and system for abdomen soft tissue nuclear magnetism image |
Non-Patent Citations (4)
| Title |
|---|
| A Road Detection Algorithm by Boosting Using Feature Combination;SHA Yun 等;《Intelligent Vehicles Symposium》;20070615;第364-368页 * |
| An Experimental Comparison of Min-Cut/Max-Flow Algorithms for Energy Minimization in Vision;Yuri Boykov 等;《PATTERN ANALYSIS AND MACHINE INTELLIGENCE》;20040930;第26卷(第9期);第1124-1137页 * |
| Image Segmentation: A Survey of Graph-cut Methods;Faliu Yi 等;《Systems and Informatics》;20120520;第1937页右栏第1-2段、第1940页右栏第1-3段 * |
| Road Detection and Classification in Urban Environments Using Conditional Random Field Models;Jyun-Fan Tsai 等;《Intelligent Transportation Systems Conference》;20060920;第963-967页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104091344A (en) | 2014-10-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977812B (en) | Vehicle-mounted video target detection method based on deep learning | |
| CN110007675B (en) | Vehicle automatic driving decision-making system based on driving situation map and training set preparation method based on unmanned aerial vehicle | |
| CN106599773B (en) | Deep learning image identification method and system for intelligent driving and terminal equipment | |
| Kühnl et al. | Monocular road segmentation using slow feature analysis | |
| CN107862261A (en) | Image people counting method based on multiple dimensioned convolutional neural networks | |
| CN105844257A (en) | Early warning system based on machine vision driving-in-fog road denoter missing and early warning method | |
| CN104715244A (en) | Multi-viewing-angle face detection method based on skin color segmentation and machine learning | |
| CN111045422A (en) | Control method for automatically driving and importing 'machine intelligence acquisition' model | |
| CN105893951A (en) | Multidimensional non-wearable type traffic police gesture identification method and system for driverless vehicles | |
| CN105930800A (en) | Lane line detection method and device | |
| CN102509098A (en) | Fisheye image vehicle identification method | |
| CN110414418A (en) | A kind of Approach for road detection of image-lidar image data Multiscale Fusion | |
| CN104091344B (en) | Road dividing method | |
| CN111881802B (en) | Traffic police gesture recognition method based on double-branch space-time graph convolutional network | |
| CN104036246A (en) | Lane line positioning method based on multi-feature fusion and polymorphism mean value | |
| CN106599848A (en) | Depth visual feature and support vector machine-based terrain texture recognition algorithm | |
| CN111079675A (en) | Driving behavior analysis method based on target detection and target tracking | |
| CN109948433A (en) | A kind of embedded human face tracing method and device | |
| Fleyeh et al. | Traffic sign detection based on AdaBoost color segmentation and SVM classification | |
| CN105205460A (en) | Face expression feature extraction and recognition method based on maximum direction encoding | |
| CN111046710A (en) | Image extraction method for importing SDL (software development language) model | |
| CN113538357B (en) | Shadow interference resistant road surface state online detection method | |
| CN106529391A (en) | Robust speed-limit traffic sign detection and recognition method | |
| CN109886125A (en) | A kind of method and Approach for road detection constructing Road Detection model | |
| CN111038521A (en) | Method for forming automatic driving consciousness decision model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |