CN103489189A - Lane detecting and partitioning method based on traffic intersection videos - Google Patents
Lane detecting and partitioning method based on traffic intersection videos Download PDFInfo
- Publication number
- CN103489189A CN103489189A CN201310439438.5A CN201310439438A CN103489189A CN 103489189 A CN103489189 A CN 103489189A CN 201310439438 A CN201310439438 A CN 201310439438A CN 103489189 A CN103489189 A CN 103489189A
- Authority
- CN
- China
- Prior art keywords
- straight line
- video image
- cluster
- traffic intersection
- adopt
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Abstract
The invention relates to the field of computer image processing, and discloses a lane detecting and partitioning method based on traffic intersection videos. The method includes the steps that Hough conversion is used for detecting video images of a traffic intersection; multi-dimensional feature vectors are extracted; a K-means algorithm is used for achieving clustering. The lane detecting and partitioning method has the advantages that good adaptability, practicability and stability are achieved, the anti-interference capacity is good, operation steps and computation steps are simplified, consumption of computing resources is reduced, besides, a stable detection result can be further obtained, and the lane detecting and partitioning method is adaptive to lane analytical application of various intelligent transportation systems and has high application value.
Description
Technical field
The present invention relates to the Computer Image Processing field, particularly a kind of lane detection and dividing method based on the traffic intersection video.
Background technology
Along with the development of computer technology and communication, the intelligent transportation system based on video technique has become the important research field of computer vision technique application.Intelligent transportation system to current traffic programme, transportation, management system, makes traffic system become the integral body of an efficient operation knowledge fusion of various advanced persons' the information processing technology, global location navigation, data communication, sensing technology, Automatic Control Theory, Digital Image Processing, computer networking technology, artificial intelligence, the subjects such as management of planning strategies for.Intelligent transportation system created a multitube reason level, comprehensive, network service, accurately, comprehensive traffic operational system at a high speed, for processing traffic problems, provide reliable technology platform to support and powerful guarantee.
Traffic information collection is the important component part of intelligent transportation system, use traditional mode can't meet current current situation to gathering transport information, the traffic information collection based on video has obtained development rapidly under this growing demand.This system can be extracted Useful Information from relevant traffic video, when the installation and maintenance video monitoring equipment, need not destroy road surface, more need not close track, can not affect normal traffic order.The most key is utilize this system not only can obtain easily each highway section at each transport information constantly, facilitate traffic monitoring, and it can also gather how valuable transport information.
Traffic information collection based on video, what at first will solve is exactly the detection in track.Because in the traffic monitoring scope, the phenomenon that multilane coexists is very general usually, track cut apart normally being arranged manually, demarcate the affiliated area in different tracks.Cut apart although the method for this artificial demarcation has solved track, reduced the applicability that traffic congestion detects, therefore automatic lane detection is very important for the detection of traffic congestion state.The automatic detection in track mainly contains the method for three kinds of main flows.First method is to utilize the calibration technique of camera, in the situation that known camera parameter, road pavement is estimated.Second method is to use the track of vehicle to carry out automatic track to cut apart, and to track analyzes of vehicle, can not be cut apart the track of complexity comparatively.The third method is to utilize the visual signature in track, because track is different from natural forms on visual characteristic, its color and edge all have obvious characteristics, the method in the traffic video monitoring field by general application.
In utilizing the visual signature method in track, due to a kind of method that lacks stable system, in the face of complicated various environment, be difficult to obtain separately good identification, so often in conjunction with some models and hypothesis, utilize the priori of road, the combining image feature estimates track, and the mathematical model, road shape hypothesis, the road surface characteristic that common are the road image end point are supposed etc.These hypothesis and model, not only limited in the traffic video the adaptability of scene, also strengthened the difficulty of operation and the complexity of realization.A kind of stable lane detection based in the traffic intersection video and dividing method are not yet proposed so far both at home and abroad.
Summary of the invention
The present invention is directed to video lane detection that in prior art, shortage is stable and the shortcoming of dividing method, a kind of novel lane detection based on the traffic intersection video and dividing method are provided.
For achieving the above object, the present invention can take following technical proposals:
Lane detection based on the traffic intersection video and dividing method comprise following concrete step:
1) utilize Hough transfer pair traffic intersection video image to be detected, extract the straight line in video image; When carrying out the Hough conversion, add the constraint condition of straight length, described constraint condition is 1/5 of picture length, gets rid of the straight line that does not meet this constraint condition; Carrying out Hough when conversion, after the pixel by image space is mapped to parameter space, is confirmed whether to exist straight line thereby whether the counting that detects each totalizer surpasses accumulation threshold, and obtains the set of the point on straight line;
2) extraction step 1) multidimensional characteristic vectors of gained straight line, described multidimensional characteristic vectors comprise straight line slope, position, length and with the matching degree of rim detection picture;
3) utilize the K-means algorithm to realize cluster, adopt half learning by rote to be analyzed in cluster gained straight line bunch, utilize the straight line at center of cluster bunch as border, mark off track; Described half learning by rote comprises the following steps: extract the fork quantity of the road junction roadway that video image absorbs, straight line bunch is compared with the fork quantity of above-mentioned road junction roadway, determine the matching degree of straight line bunch and the fork quantity of above-mentioned road junction roadway; Described K-means algorithm comprises the following steps: adopt the method for multicenter cluster comparison to determine the number at center, adopt the K-means algorithm to realize step 2) the gained multidimensional characteristic vectors analyzed and extracted initial cluster center.
As preferably, further comprising the steps of:
Pre-treatment step: adopt Fast Median Filtering to carry out pre-service to the traffic intersection video image, move neighborhood window on video image, when the gray-scale value of the pixel that newly enters the neighborhood window is different from the gray-scale value of the pixel of new grand window, the neighborhood window is asked to intermediate value, adopt insertion sort simultaneously, in sequencer procedure, existing sequence is immobilized, thereby obtain filter effect faster;
Edge detecting step: use Canny operator or Sobel operator to carry out rim detection to video image and obtain the rim detection picture, wherein, the video image absorbed under the intense light irradiation condition is carried out to the edge enhancing, to normal care and the low light level, according to the video image absorbed under condition, utilize the Sobel operator to calculate the gray difference value of neighbor pixel.
As preferably, it is characterized in that, further comprising the steps of: that every straight line is also extracted to the high-order proper vector of the gradient space value of the color value of its extended line and border intersecting point coordinate, video image and video image as straight line; Slope is replaced with slope place straight line and horizontal angle; Described multidimensional characteristic vectors and high dimensional feature vector are carried out to that normalization; When calculating the weight of multidimensional characteristic vectors and high-order proper vector, keep the weighted value of slope and position to be greater than the weighted value that other mostly are proper vector or high-order proper vector; Adopt statistical method, select adaptively the number of cluster, and compare with cluster gained straight line bunch, by following algorithm, obtain optimum solution:
wherein, the centric quantity that n is cluster, V
pfor the proper vector of every straight line, C
nproper vector for corresponding cluster centre.
As preferably, it is characterized in that, for step 3) gained track, according to the track quantity obtained, extract the Along ent identical with track quantity as Seed Points in video image, spread in Hough conversion gained rectilinear, mark off further different tracks.
The present invention, owing to having adopted above technical scheme, has significant technique effect:
The present invention is based on the video image that traffic intersection is taken, and can obtain according to Analysis of test results the lane line at crossing, and can be partitioned into adaptively effective track.When this method is applied in intelligent transportation system, there is adaptability preferably, can avoid adopting model or suppose the restriction to crossing, go for various dissimilar crossings, also simplified operation steps simultaneously, save computational resource, improved the processing speed of intelligent transportation system.The present invention has real-time preferably, has saved in prior art operation links such as crossing being carried out to modeling, especially is adapted to the intelligent transportation system had higher requirements for real-time.In addition, result of calculation of the present invention is more stable, detects and cuts apart resulting lane information and can directly input next step application link, can organically combine together with other links of intelligent transportation system.
Embodiment
Below in conjunction with embodiment, the present invention is described in further detail.
Embodiment 1
When implementing lane detection based on the traffic intersection video image and cutting techniques, utilize video image as the entrance of processing, can adopt the picture Processing Technique of the maturation such as gauss hybrid models to be processed the video image of traffic intersection.
Lane detection based on the traffic intersection video and dividing method, its detailed process is as follows:
The first step, adopt the improvement Fast Median Filtering to carry out pre-service to video image.Wherein, because the grey scale change of view data in video image of lane line is little, can think that the lane line region has identical gray-scale value, therefore, the neighborhood window there is no need all to calculate its intermediate value at every turn while moving on entire image, only have when the gray-scale value of the gray-scale value that newly enters neighborhood window pixel and grand window pixel does not wait, just the neighborhood window is asked to intermediate value.And adopt insertion sort when gray-scale value does not wait, keep existing sequence to immobilize, thereby obtain filter effect faster.
Second step, rim detection is extracted lane information.The gray difference feature of Sobel operator reflection neighbor pixel, calculated amount is little, algorithm is simple, all better aspect edge enhancing effect, inhibition noise ability, but the edge extracted is thicker; The target of Canny operator is to find the edge detection algorithm of an optimum, and the edge of this operator extraction is complete, and the lane line detected is that single pixel is wide, but the algorithm complexity, calculated amount is larger.Consider real-time and rim detection effect, the edge that carries out the intense light irradiation image with the Canny algorithm strengthens, and shines image for normal illumination and the low light level, adopts the Sobel algorithm.
The 3rd step, utilize improved Hough change detection straight line.Thereby traditional Hough conversion is to add up the counting of each totalizer whether to be greater than accumulation threshold and to determine whether to exist straight line at image space after the mapping of parameter space all completes, and the set of the point on the acquisition respective straight, improved Hough conversion adopts: after the point in image space being mapped to parameter space, all can check whether the counting of each totalizer has surpassed accumulation threshold at every turn, the maximum probability that the longest like this straight line is arrived by earliest detection, thus the operand of detection of straight lines reduced.
The 4th step, the multidimensional characteristic vectors cluster of extraction straight line.By top operation, what we obtained is much lane line marginal distribution length, the different straight line of slope, how thereby these straight lines are effectively sorted out and analyzed the difficult problem that lane line is the independent widespread use based on the video detection of traffic flow system of restriction always, the present invention adopts following multidimensional characteristic vectors clustering method, analyzes different lane informations:
1) every straight line is extracted to the intersection point on slope, position, length, extended line and border, former figure color value, former figure gradient space value, as the high dimensional feature vector of every straight line.
2), because slope variation is extremely inhomogeneous, we are translated into horizontal angle and calculate.Each dimension is carried out normalization, and guarantees that the important dimension weights such as slope, position are larger.
3) for the number difference of different road junction roadway lines, we adopt the method for statistics, can carry out in advance complicate statistics, select the number of cluster according to dynamic range, then in conjunction with the cluster result (result that refers to different clusters number.Be the data training stage at present, still in the preliminary preparation stage) analyze relatively, thus find optimum solution (dynamic deviation of cluster, select optimum number, optimum center):
here the centric quantity that n is cluster, V
pfor the proper vector of each straight line, C
nproper vector for corresponding cluster centre.
4) in conjunction with top result, adopt improved K-means algorithm to carry out cluster to the gained straight line, to choosing of the cluster centre point with initial, adopt the K-means algorithm to be analyzed multidimensional characteristic vectors, extract initial cluster center, avoid algorithm to be absorbed in local optimum.
The 5th step, utilize the straight line at center of cluster as border, marks off final track.Coordinate system according to former figure, come zoning, because lane line is varied, be difficult to directly find unified track relation, we are by the track quantity of upper surface analysis, at the picture stage identical Along ent of quantity of picking up the car, it is Seed Points, and spread in the straight line picture, the diffusion process here adopts algorithm of region growing, adopts by choosing a Seed Points, and to the mode of external diffusion, thereby finally mark off different tracks.
In a word, the foregoing is only preferred embodiment of the present invention, all equalizations of doing according to the present patent application the scope of the claims change and modify, and all should belong to the covering scope of patent of the present invention.
Claims (4)
1. lane detection and the dividing method based on the traffic intersection video, is characterized in that, comprises following concrete step:
1) utilize Hough transfer pair traffic intersection video image to be detected, extract the straight line in video image; When carrying out the Hough conversion, add the constraint condition of straight length, described constraint condition is 1/5 of picture length, gets rid of the straight line that does not meet this constraint condition; Carrying out Hough when conversion, after the pixel by image space is mapped to parameter space, is confirmed whether to exist straight line thereby whether the counting that detects each totalizer surpasses accumulation threshold, and obtains the set of the point on straight line;
2) extraction step 1) multidimensional characteristic vectors of gained straight line, described multidimensional characteristic vectors comprise straight line slope, position, length and with the matching degree of rim detection picture;
3) utilize the K-means algorithm to realize cluster, adopt half learning by rote to be analyzed in cluster gained straight line bunch, utilize the straight line at center of cluster bunch as border, mark off track; Described half learning by rote comprises the following steps: extract the fork quantity of the road junction roadway that video image absorbs, straight line bunch is compared with the fork quantity of above-mentioned road junction roadway, determine the matching degree of straight line bunch and the fork quantity of above-mentioned road junction roadway; Described K-means algorithm comprises the following steps: adopt the method for multicenter cluster comparison to determine the number at center, adopt the K-means algorithm to realize step 2) the gained multidimensional characteristic vectors analyzed and extracted initial cluster center.
2. lane detection and the dividing method based on the traffic intersection video according to claim 1, is characterized in that, further comprising the steps of:
Pre-treatment step: adopt Fast Median Filtering to carry out pre-service to the traffic intersection video image, move neighborhood window on video image, when the gray-scale value of the pixel that newly enters the neighborhood window is different from the gray-scale value of the pixel of new grand window, the neighborhood window is asked to intermediate value, adopt insertion sort simultaneously, in sequencer procedure, existing sequence is immobilized, thereby obtain filter effect faster;
Edge detecting step: use Canny operator or Sobel operator to carry out rim detection to video image and obtain the rim detection picture, wherein, the video image absorbed under the intense light irradiation condition is carried out to the edge enhancing, to normal care and the low light level, according to the video image absorbed under condition, utilize the Sobel operator to calculate the gray difference value of neighbor pixel.
3. lane detection and the dividing method based on the traffic intersection video according to claim 1, it is characterized in that, it is characterized in that, further comprising the steps of: that every straight line is also extracted to the high-order proper vector of the gradient space value of the color value of its extended line and border intersecting point coordinate, video image and video image as straight line; Slope is replaced with slope place straight line and horizontal angle; Described multidimensional characteristic vectors and high dimensional feature vector are carried out to that normalization; When calculating the weight of multidimensional characteristic vectors and high-order proper vector, keep the weighted value of slope and position to be greater than the weighted value that other mostly are proper vector or high-order proper vector; Adopt statistical method, select adaptively the number of cluster, and compare with cluster gained straight line bunch, by following algorithm, obtain optimum solution:
wherein, the centric quantity that n is cluster, V
pfor the proper vector of every straight line, C
nproper vector for corresponding cluster centre.
4. lane detection and the dividing method based on the traffic intersection video according to claim 1, it is characterized in that, it is characterized in that, for step 3) gained track, according to the track quantity obtained, extract the Along ent identical with track quantity as Seed Points in video image, spread in Hough conversion gained rectilinear, mark off further different tracks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310439438.5A CN103489189B (en) | 2013-09-24 | 2013-09-24 | Lane detection based on traffic intersection video and dividing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310439438.5A CN103489189B (en) | 2013-09-24 | 2013-09-24 | Lane detection based on traffic intersection video and dividing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103489189A true CN103489189A (en) | 2014-01-01 |
| CN103489189B CN103489189B (en) | 2016-05-18 |
Family
ID=49829387
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310439438.5A Active CN103489189B (en) | 2013-09-24 | 2013-09-24 | Lane detection based on traffic intersection video and dividing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103489189B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184240A (en) * | 2015-08-27 | 2015-12-23 | 广西师范学院 | Scan line clustering-based security video road automatic identification algorithm |
| CN107220976A (en) * | 2017-05-17 | 2017-09-29 | 南京航空航天大学 | A kind of highway localization method for highway map picture of taking photo by plane |
| CN107977608A (en) * | 2017-11-20 | 2018-05-01 | 陕西土豆数据科技有限公司 | A kind of method applied to the extraction of highway video image road area |
| CN109871752A (en) * | 2019-01-04 | 2019-06-11 | 北京航空航天大学 | A method of lane line is extracted based on monitor video detection wagon flow |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101567086A (en) * | 2009-06-03 | 2009-10-28 | 北京中星微电子有限公司 | Method of lane line detection and equipment thereof |
| JP2013101534A (en) * | 2011-11-09 | 2013-05-23 | Ricoh Co Ltd | Vehicle driving support system |
-
2013
- 2013-09-24 CN CN201310439438.5A patent/CN103489189B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101567086A (en) * | 2009-06-03 | 2009-10-28 | 北京中星微电子有限公司 | Method of lane line detection and equipment thereof |
| JP2013101534A (en) * | 2011-11-09 | 2013-05-23 | Ricoh Co Ltd | Vehicle driving support system |
Non-Patent Citations (2)
| Title |
|---|
| PANGYU JEONG等: "Efficient and Robust Classification Method Using Combined Feature Vector for Lane Detection", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| 杨喜宁 等: "基于改进Hough变换的车道线检测技术", 《计算机测量与控制》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184240A (en) * | 2015-08-27 | 2015-12-23 | 广西师范学院 | Scan line clustering-based security video road automatic identification algorithm |
| CN105184240B (en) * | 2015-08-27 | 2018-05-29 | 广西师范学院 | Security protection video road automatic identification algorithm based on scan line cluster |
| CN107220976A (en) * | 2017-05-17 | 2017-09-29 | 南京航空航天大学 | A kind of highway localization method for highway map picture of taking photo by plane |
| CN107220976B (en) * | 2017-05-17 | 2020-11-20 | 南京航空航天大学 | Highway positioning method for aerial highway image |
| CN107977608A (en) * | 2017-11-20 | 2018-05-01 | 陕西土豆数据科技有限公司 | A kind of method applied to the extraction of highway video image road area |
| CN107977608B (en) * | 2017-11-20 | 2021-09-03 | 土豆数据科技集团有限公司 | Method for extracting road area of highway video image |
| CN109871752A (en) * | 2019-01-04 | 2019-06-11 | 北京航空航天大学 | A method of lane line is extracted based on monitor video detection wagon flow |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103489189B (en) | 2016-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103383733B (en) | A kind of track based on half machine learning video detecting method | |
| US9224046B2 (en) | Multi-view object detection using appearance model transfer from similar scenes | |
| CN104303193B (en) | Target classification based on cluster | |
| US7869935B2 (en) | Method and system for detecting traffic information | |
| US8970701B2 (en) | System and method for predicting vehicle location | |
| CN103258203B (en) | The center line of road extraction method of remote sensing image | |
| CN103714538B (en) | road edge detection method, device and vehicle | |
| CN103150903B (en) | Video vehicle detection method for adaptive learning | |
| CN102819764B (en) | Method for counting pedestrian flow from multiple views under complex scene of traffic junction | |
| CN104951775A (en) | Video technology based secure and smart recognition method for railway crossing protection zone | |
| CN102385803B (en) | All-weather urban vehicle tracking and counting method based on video monitoring | |
| CN104050481B (en) | Multi-template infrared image real-time pedestrian detection method combining contour feature and gray level | |
| CN104112370A (en) | Monitoring image based intelligent parking lot parking place identification method and system | |
| CN103871079A (en) | Vehicle tracking method based on machine learning and optical flow | |
| CN101976504B (en) | Multi-vehicle video tracking method based on color space information | |
| CN102819952A (en) | Method for detecting illegal lane change of vehicle based on video detection technique | |
| CN104268583A (en) | Pedestrian re-recognition method and system based on color area features | |
| CN103206957B (en) | The lane detection and tracking method of vehicular autonomous navigation | |
| CN109800773B (en) | Cross-country road surface extraction method based on three-dimensional laser radar | |
| CN104134067A (en) | Road vehicle monitoring system based on intelligent visual Internet of Things | |
| US10685443B2 (en) | Cloud detection using images | |
| CN110633678B (en) | Quick and efficient vehicle flow calculation method based on video image | |
| CN103489189B (en) | Lane detection based on traffic intersection video and dividing method | |
| CN110379168A (en) | A kind of vehicular traffic information acquisition method based on Mask R-CNN | |
| CN103325259A (en) | Illegal parking detection method based on multi-core synchronization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |