CN101364347A - Detection method for vehicle delay control on crossing based on video - Google Patents

Abstract

The invention discloses a video-based detection method for the vehicle control delay at a road intersection. The method comprises the following steps: arranging cameras around the road intersection to obtain a video record of a certain approach vehicle queue at the road intersection; performing the global calibration and the coordinate transformation of the video record to obtain the corresponding relationship between the image coordinate and the three-dimensional global coordinate; performing the background update, the image binarization and the plaque analysis of the video record; setting distal and proximal vehicle detection zones for the triggering and detection of the vehicle; recording the time when the detected vehicle drives into the distal vehicle detection zone and the time when the vehicle passes through the stop line by tracking the detected vehicle to obtain the effective time when the vehicle drives through the road intersection; and subtracting the free driving time when the vehicle passes through the intersection to obtain the control delay as key traffic parameter when the vehicle passes through the intersection. The detection method can provide a signal controller with the stable, real-time and reliable information such as average vehicle control delay.

Description

Crossing vehicle based on video is controlled the detection method of incuring loss through delay

Technical field

The invention belongs to transport information and control field, the acquisition methods that relates to a kind of crossing traffic parameter and signal controlling evaluation index is more specifically said so and is utilized video detection technology to obtain the method that this important traffic parameter and core evaluation index are incured loss through delay in crossing vehicle control automatically.

Background technology

Signalized crossing vehicle control is incured loss through delay and is meant because vehicle is subjected to the loss of time that influence that intersection signal controls causes, it has reflected that not only the driver is uncomfortable and has been obstructed degree and oil consumption and running time loss, the rationality that has also reflected crossing traffic design, signal controlling scheme, be the important indicator of estimating crossing degree of congestion, service level, can provide the important quantitative foundation for crossing means of transportation and the improvement of signal controlling scheme.

But to so far, traffic parameters such as crossing vehicle delay mainly obtain by the artificial observation method, but artificial observation need expend lot of manpower and material resources, and can't satisfy the demand that the signal controlling scheme is adjusted optimization in real time, in order to overcome all drawbacks of artificial observation traffic parameter, people begin attempting utilizing sensor technology, the microelectric technique and the information processing technology detect the traffic parameter of crossing automatically, can be divided into the magnetic induction detection technique according to the know-why that adopts, ripple is detection technique and three kinds of methods of video detection technology frequently, but magnetic induction detection technique and ripple detection technique frequently can only directly detect the volume of traffic, parameters such as occupation rate can't be obtained the core parameters such as delay of vehicle by the crossing.

Video detection technology is a kind of in conjunction with video image and mode identification technology and be applied to the emerging technology of field of traffic.Video detection technology is by following the tracks of the vehicle in the certain limit on all tracks of certain direction of road, not only can provide the magnitude of traffic flow and road speed, and can obtain these crossing operation characteristic parameters such as density, lane changing, acceleration, queuing and parking that the conventional sense technology can't be obtained simultaneously, especially incuring loss through delay for the crucial operational factor of obtaining the crossing provides possibility.

Find through literature search prior art, DETECTION OF TRAFFIC PARAMETERS based on video becomes the development trend that traffic parameter detects automatically, many scholars have carried out relevant research, delivered a large amount of achievements in research, but up to the present, the video traffic parameter detecting mainly is confined to the magnitude of traffic flow, average velocity and lane occupancy ratio, the detection of the conventional traffic parameter in crossings such as queue length, utilize Model Calculation to obtain the core evaluation index of this intersection signal control of vehicle delay indirectly then, utilize video detection technology directly very rare the research that intersection delay detects automatically.People such as Ran Qiwu deliver the article of " detecting based on the instantaneous stop delay of the vehicle of video " in 2007 the 6th phases " XI AN JIAOTONG UNIVERSITY Subject Index ", a kind of detection method of stop delay has been proposed, this square tube is crossed the vehicle number that automatic identification video is used videotape to record and stopped before the crossing in each frame, calculate the instantaneous stop delay of stopped vehicle before each frame crossing, the instantaneous stop delay that adds up all frames then promptly obtains observing the total amount of the instantaneous stop delay of period vehicle, but what this method obtained is the stop delay of crossing vehicle, not comprising stops vehicle deceleration the deceleration cause is incured loss through delay and accelerate to the acceleration that cruising speed caused by stationary state incurs loss through delay, in the time of can not accurately reflecting vehicle by the crossing because the loss of time that signal controlling caused.

Summary of the invention

Technical matters to be solved by this invention is can not directly detect the deficiency that vehicle is incured loss through delay by the control of crossing at prior art, adopt Video Detection and target following technology, automatically obtain the parameters such as control delay of signalized crossing, a kind of detection method of incuring loss through delay based on the crossing vehicle control of video is provided.

The technical scheme of technical solution problem of the present invention is as follows:

A kind of detection method of incuring loss through delay based on the crossing vehicle control of video is characterized in that, comprises the steps:

(1), hardware system makes up and the video picture recording obtains: select equipment such as commanding elevation installation camera, switch, video server near the crossing, and with electric wire, optical cable or wireless they are coupled together;

(2), global calibration and coordinate conversion: target area in the selecting video picture recording, carry out the coordinate transform of former figure and vertical view, obtain the corresponding relation of image coordinate and three-dimensional world coordinate;

(3), context update: obtain initial background, realize the context update of vertical view;

(4), image binaryzation and patch analysis: create the corresponding gray-scale map of vertical view, obtain the difference diagram of background and current vertical view, and the binaryzation of difference diagram is realized on ground, divided lane subregion, the binary map that obtains is done patch analysis and cluster, obtain big or small length and width and meet the patch information that vehicle shape requires;

(5), vehicle triggers and detection: two surveyed areas of far-end and near-end are set, and the far-end surveyed area is arranged on the upstream, track away from maximum queuing place of vehicle of camera, and the near-end surveyed area is arranged near the track part near camera of stop line;

(6), vehicle tracking;

(7), traffic parameter extracts: vehicle is the control delay that vehicle passes through the crossing by the free running time that deducts vehicle actual consuming time of crossing.

By above disclosed technical scheme as can be known, the present invention is owing to realize tracking control of full process, system writes down the moment that each vehicle enters far-end surveyed area and stop line automatically, pass through the actual consuming time of crossing thereby obtain vehicle, vehicle promptly obtains the control delay of vehicle by the crossing by the free running time that deducts vehicle actual consuming time of crossing.System can also obtain conventional traffic parameters such as the volume of traffic, various vehicle ratio, average velocity and lane occupancy ratio, queue length simultaneously automatically.The present invention has formed a kind of crossing traffic information collection, analysis and disposal system of robotization, information such as stable, timely, reliably average vehicle control delay are provided to signal controller, make that whole whistle control system is with clearly defined objective, have solid detection basis, the system that reaches is controlled, the purpose of optimization.

Description of drawings

Fig. 1 a is a photography conversion perspective view;

Fig. 1 b is the mapping graph of overlooking of photography conversion perspective view;

Fig. 2 is that the detection synoptic diagram is incured loss through delay in control;

Embodiment

Below in conjunction with accompanying drawing embodiments of the invention are elaborated: present embodiment is being to implement under the prerequisite with the technical solution of the present invention, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.

Need experience following process based on the detection that the crossing vehicle control of video is incured loss through delay: hardware system makes up with the video picture recording and obtains, demarcation of video picture recording and steps such as coordinate conversion, context update, image binaryzation, patch analysis, vehicle triggering and detection, vehicle tracking, traffic parameter extraction, and specifically implementation step is as follows:

1, hardware system makes up with the video picture recording and obtains:

Near the crossing, select a commanding elevation that equipment such as camera, switch, video server are installed, and with electric wire, optical cable or wireless they are coupled together, obtain the video record of crossing inlet road vehicle queue, and video record is transferred to video server by switch.

2, global calibration and coordinate conversion:

The target area that will detect in the selecting video picture recording, it is set to area-of-interest (bicycle road or multilane), utilize the calibration algorithm of vanishing point and the constant theorem of double ratio to realize the coordinate transform of former figure area-of-interest and vertical view, obtain the corresponding relation of image coordinate and three-dimensional world coordinate, create the vertical view in research track, all follow-up processing are all carried out overlooking on the mapping graph, can significantly reduce calculated amount thus.

Based on the principle of the conversion of settling down of vanishing point as shown in Figure 1, establish that rectangle OABC is wide to be W, longly be L, it is projected as quadrilateral oabc through the photography conversion, d be on the oabc of plane more arbitrarily, the coordinate after known o, a, b, c, the d photography conversion is asked the coordinate of corresponding point D on OABC before the d conversion.Under perspective projection transformation, the image of space line still is a straight line, and one group of parallel lines in space will be projected as crossing straight line, and intersect at unique a bit, i.e. vanishing point.Vanishing point generally is positioned at outside the image, even at infinity.All vanishing points on the georeferencing plane (and the plane that parallels with it) all are located on the same line, and are referred to as blanking line.Vanishing point and blanking line are implied with the directional information on space line and plane, owing to contain the road traffic marking that is parallel to each other in the traffic video, the length of traffic marking and its spacing field survey in advance obtain, in the global calibration process, can make full use of these information, and do not need to know camera parameters and with the transformation relation of scene.

Since AB ∥ OC, corresponding vanishing point after releasing oc and the intersection point v1 of ab being all straight line parallel with AB photography conversion; In like manner, oa and corresponding vanishing point after the intersection point v2 of bc also is all straight line conversion parallel with OA.Can obtain v according to vanishing point ₂The intersection point m of d and ab is the corresponding point of D point after projection M photography conversion on the AB; v ₁The intersection point n of d and oa is the corresponding point of the projection N of D point on OA.

According to the double ratio unchangeability, have:

(oa；nv ₂)＝(OA；NV ₂)(bm；av ₁)＝(BM；AV ₁)

Can obtain formula:

$(\mathrm{oa};{\mathrm{nv}}_2)=\frac{\stackrel{\‾}{\mathrm{on}}\×\stackrel{\‾}{a{v}_2}}{\stackrel{\‾}{\mathrm{an}}\×\stackrel{\‾}{o{v}_2}}=\frac{\stackrel{\‾}{\mathrm{ON}}\×\stackrel{\‾}{A{V}_2}}{\stackrel{\‾}{\mathrm{AN}}\×\stackrel{\‾}{O{V}_2}}=\frac{\stackrel{\‾}{\mathrm{ON}}}{\stackrel{\‾}{\mathrm{AN}}}=(\mathrm{OA};{\mathrm{NV}}_2)$

$\stackrel{\‾}{\mathrm{ON}}=\stackrel{\‾}{\mathrm{AN}}\×\frac{\stackrel{\‾}{\mathrm{on}}\×\stackrel{\‾}{{\mathrm{av}}_2}}{\stackrel{\‾}{\mathrm{an}}\×\stackrel{\‾}{o{v}_2}}$

$(\mathrm{bm};{\mathrm{av}}_1)=\frac{\stackrel{\&OverBar;}{\mathrm{ba}}\&times;\stackrel{\&OverBar;}{m{v}_1}}{\stackrel{\&OverBar;}{\mathrm{ma}}\&times;\stackrel{\&OverBar;}{b{v}_1}}=\frac{\stackrel{\&OverBar;}{\mathrm{BA}}\&times;\stackrel{\&OverBar;}{\mathrm{<figure-callout\; id="1"\; label="M\; V"\; filenames="A200810200001E00092.png"\; state="\{\{state\}\}">M</figure-callout>}{V}_{}{}_1}}{\stackrel{\&OverBar;}{\mathrm{MA}}\&times;\stackrel{\&OverBar;}{\mathrm{<figure-callout\; id="1"\; label="B\; V"\; filenames="A200810200001E00092.png"\; state="\{\{state\}\}">B</figure-callout>}{V}_{}{}_1}}=\frac{\stackrel{\&OverBar;}{\mathrm{BA}}}{\stackrel{\&OverBar;}{\mathrm{MA}}}=(\mathrm{BM};{\mathrm{AV}}_1)$

$\stackrel{\&OverBar;}{\mathrm{MA}}=\stackrel{\&OverBar;}{\mathrm{BA}}\&times;\frac{\stackrel{\&OverBar;}{\mathrm{ma}}\&times;\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="bv"\; filenames="A200810200001E00092.png"\; state="\{\{state\}\}">bv</figure-callout>}}_1}}{\stackrel{\&OverBar;}{\mathrm{ba}}\&times;\stackrel{\&OverBar;}{m{v}_1}}$

3, context update:

At signalized crossing, red interval often has vehicle queue, and the road surface is blocked for a long time by vehicle, and in such cases, general context update algorithm often is updated to vehicle color in the background gradually, thereby causes background contamination.In order to overcome this phenomenon, the present invention utilizes adaptive median filter AMF algorithm to obtain initial background, in conjunction with the context update of motion mask in the Surendra algorithm and AMF algorithm realization vertical view.Make the motion mask according to the zone that has had vehicle in the track and candidate's vehicle region (patch zone), context update is not carried out in the zone in this mask, but not masks area then can be realized context update rapidly according to the AMF algorithm.

The AMF basic idea is that if the pixel value of new incoming frame is bigger than the pixel value of the correspondence position in the background of estimating, then the pixel value of estimated background increases 1, if littler than the pixel value of corresponding position, then subtracts 1.This background estimating will finally converge on such value, that is: general input pixel value is bigger than this value, and the pixel value of half is littler than this value, and this is worth intermediate value just.

Final context update formula may be defined as

$B_i(x,y)=\left\{\begin{array}{cc}{B}_{i-1}(x,y)& \mathrm{ifB}{W}_i(x,y)=1\mathrm{or}{I}_i(x,y)=B_{i-1}(x,y)\\ B_{i-1}(x,y)\&PlusMinus;1& \mathrm{if}{\mathrm{BW}}_i(x,y)=0,I_i(x,y)>B_{i-1}(x,y),+1;I_i(x,y)<B_{i-1}(x,y),-1\end{array}\right.$

B wherein _i(x is that current i frame background is (x, the gray-scale value of y) locating, B at coordinate y) _I-1(x is that the background of i-1 frame is (x, the gray-scale value of y) locating, I at coordinate y) _i(x is that current i two field picture is (x, the gray-scale value of y) locating, BW at coordinate y) _i(x is that (1 for being in masks area for x, the value of y) locating, and 0 for being in non-masks area for coordinate in the motion mask artwork y).

4, image binaryzation and patch analysis:

Convert the colored vertical view that obtains to gray-scale map and can reduce computational complexity significantly, when needs are handled full figure, often at be gray-scale map, only when vehicle detection and tracking, just can use the chromatic information of vehicle region, image binaryzation then is the basis of extracting information of vehicles, its operation to as if the gray-scale map of the difference of prospect and background.Image binaryzation provides resource for ensuing patch analysis, the part foundation when also being context update simultaneously.

The influence of vehicle color between the different tracks in the binaryzation process, the present invention is divided into a plurality of parts with vertical view by the track, and each track is divided into three parts, front-end detection zone and near-end surveyed area respectively are a part, track between two surveyed areas is separately as a part, the part of ultimate demand binaryzation is 3*N, wherein N is the number of lanes in the vertical view, all parts are adopted Otsu method binaryzation respectively, so not only can improve the accuracy rate of Threshold Segmentation and the accuracy that each surveyed area detects vehicle greatly, also can improve the tracking effect of two tracing areas between the surveyed area.

Behind the context update, prospect is split by image partition method, being reflected on the binary map is connected region one by one, be referred to as patch, the patch analysis need be extracted the essential information of all patches in the binary map, again according to the position and the colouring information of patch, and polymerization or cut apart patch, and the last patch that does not meet the vehicle size shape of deleting, finish the patch cluster.

5, vehicle triggers and detects:

By the result who obtains after the above-mentioned patch cluster, we can obtain candidate's vehicle region, i.e. the zone of patch after the cluster, and each the candidate's vehicle in the monitoring and detection zone then can be realized the detection and the triggering of actual vehicle.In Video Detection, vehicle is at first to enter the far-end surveyed area, because perspective relation, vehicle is little at near-end at the far-end ratio among the former figure, the information that self provides is also few, though vehicle size is constant substantially in vertical view, unlike proximal end region, the one part of pixel value of remote area is that interpolation gets during by conversion.So easily cause the vehicle omission at upstream region.Simultaneously, the tracing area between two surveyed areas, detected vehicle also may cause for various reasons and follow the tracks of failure.The number of vehicles of process is to influence the key factor that mean delay calculates in the track, if number of vehicles is inaccurate, the calculating of mean delay also can be introduced suitable error.In order to improve the precision of vehicle detection, reduce the loss of vehicle, the present invention is provided with far-end and two surveyed areas of near-end, the far-end surveyed area is arranged on the upstream, track away from maximum queuing place of vehicle of camera, the near-end surveyed area is arranged near the track part near camera of stop line, when mistake in far-end surveyed area omission vehicle or the tracing process is deleted vehicle, can be triggered once more at the near-end surveyed area and detect, can estimate by car information before and after it in the be triggered relevant information of detected vehicle of near-end, thereby reduced the loss of vehicle, improved the precision that detects.

6, vehicle tracking:

In traffic video monitoring system, target following plays crucial effects.Behind the context update, by modules such as image binaryzation and patch analysis, vehicle detection and triggerings, find the position and the size of the vehicle that newly enters surveyed area at last, and go out with the square frame frame, to carry out target following with that, to obtain to comprise vehicle instantaneous velocity, average velocity, road occupies most of basic transport information such as situation, and provides vehicle and road condition information accurately for abnormality detection.Consider the real-time requirement, under the situation of not blocking, use patch to follow the tracks of, back Mean shift track algorithm occurs blocking.The robustness of supervisory system real-time follow-up has been strengthened in the use of Mean shift method, has reduced the Loss Rate of back target to occur blocking and detect chaotic.

The patch tracking promptly when a new frame arrives, in the position that a new frame vehicle may occur, is done histogram near the patch this position according to the estimation of previous frame vehicle position, and does coupling with the histogram of vehicle.If there is the patch of coupling, then upgrade the position and the histogram of vehicle.When calculating the histogram of patch, for 24RGB figure, in order to reduce computation complexity, each color channel is quantified as 8 values, therefore always have 8 * 8 * 8=512 value in the histogram, it is original 256 * 256 * 256 that calculated amount is less than far away, but precision is constant substantially to the influence of effect.Because the influence of vehicle self color and Threshold Segmentation may not have patch near the position that vehicle is estimated, perhaps there is not the patch of coupling, at this moment, patch is followed the tracks of failure.Red interval, the situation of blocking can appear in the vehicle parking queuing, and this was sticked together with the patch that different vehicle is mated respectively originally, was rendered as a patch, and patch was followed the tracks of and also can be failed this moment.

When following the tracks of failure, then adopts patch the meanshift tracking, it does not rely on patch, it is a kind of self-adaption gradient algorithm based on Density Estimator, utilize Bayes's coefficient to weigh similarity between object and the template based on picture element color and position, so that in image, find out the destination object the most similar, has certain anti-ability of blocking to template.

7, traffic parameter extracts:

The synoptic diagram that control incur loss through delay to detect as shown in Figure 2, vehicle enters the far-end surveyed area and is triggered and detects, system notes the moment (the observation data t among Fig. 2 that this car enters surveyed area automatically ₁), by all-the-way tracking, when vehicle passed through stop line, system noted the moment (the observation data t among Fig. 2 that this car leaves stop line simultaneously ₂), then vehicle is by the real time T of crossing _aFor:

T _a=t ₂-t ₁(formula 1)

Normal vehicle operation is by the time T of crossing _bIn the time of should be by green light directly the vehicle by stop line record, vehicle is incured loss through delay d by the control of crossing and is:

D=T _a-T _b(formula 2)

Claims (5)

1. a detection method of incuring loss through delay based on the crossing vehicle control of video is characterized in that, comprises the steps:

(1), hardware system makes up and the video picture recording obtains: select equipment such as commanding elevation installation camera, switch, video server near the crossing, and with electric wire, optical cable or wireless they are coupled together;

(2), global calibration and coordinate conversion: target area in the selecting video picture recording, carry out the coordinate transform of former figure and vertical view, obtain the corresponding relation of image coordinate and three-dimensional world coordinate;

(3), context update: obtain initial background, realize the context update of vertical view;

(4), image binaryzation and patch analysis: create the corresponding gray-scale map of vertical view, obtain the difference diagram of background and current vertical view, and the binaryzation of difference diagram is realized on ground, divided lane subregion, the binary map that obtains is done patch analysis and cluster, obtain big or small length and width and meet the patch information that vehicle shape requires;

(5), vehicle triggers and detection: two surveyed areas of far-end and near-end are set, and the far-end surveyed area is arranged on the upstream, track away from maximum queuing place of vehicle of camera, and the near-end surveyed area is arranged near the track part near camera of stop line;

(6), vehicle tracking;

(7), traffic parameter extracts: vehicle is the control delay that vehicle passes through the crossing by the free running time that deducts vehicle actual consuming time of crossing.

2. the detection method of incuring loss through delay based on the crossing vehicle control of video according to claim 1, it is characterized in that, in the said step (2), the calibration algorithm of employing vanishing point and the constant theorem of double ratio is realized the coordinate transform of former figure area-of-interest and vertical view.

3. the detection method of incuring loss through delay based on the crossing vehicle control of video according to claim 1, it is characterized in that, in the said step (3), utilize adaptive median filter AMF algorithm to obtain initial background, in conjunction with the context update of motion mask in the Surendra algorithm and AMF algorithm realization vertical view.

4. the detection method of incuring loss through delay based on the crossing vehicle control of video according to claim 1 is characterized in that, in the said step (4), adopts ground, otsu algorithm divided lane subregion to realize the binaryzation of difference diagram.

5. the detection method of incuring loss through delay based on the crossing vehicle control of video according to claim 1, it is characterized in that, in the said step (6), detected vehicle is adopted the method that patch is followed the tracks of and the tracking of meanshift method combines, realize that vehicle is from entering upstream far-end surveyed area, until the tracking control of full process that leaves stop line.

Images