CN103903445A - Vehicle queuing length detection method and system based on video - Google Patents

Abstract

The invention provides a vehicle queuing length detection method and system based on a video in the intelligent transportation system. The method and system can provide a certain digital basis to relieve and solve traffic jams. The method relates to a system initialization unit used for extracting information and establishing initialization parameters, an environment classification unit used for environment classification and switching and guaranteeing the accuracy of the system in different weathers and environments, a vehicle feature extracting unit for extracting prominent vehicle features in the current environment, and a vehicle queuing detection unit for finding out the tail of a queue and calculating the queuing length. According to the method and system, a camera model is established by means of standard pavement markers without the assistance of a calibration board, and the parameters of the model can be adjusted in real time to reduce the errors caused by tiny offset of a camera; the tail of the queue is looked for by controlling the movement of an interested area, the vacant phenomenon during vehicle queuing can be dealt with, real-time requirements of video monitoring are met, the detection efficiency and accuracy of the systemare improved, the application range is wide, and stability and reliability are high.

Description

A kind of vehicle queue length detection method and system based on video

Technical field

The present invention relates to intelligent transportation field, particularly in the system such as road traffic control and expressway tol lcollection, utilize the system that detects the vehicle characteristics of area-of-interest and state and control area-of-interest and move to vehicle queue's tail of the queue.

Background technology

Along with developing rapidly of transportation, it is more and more serious that traffic congestion phenomenon has become.Especially the jam before freeway toll station and city traffic signal lamp is very outstanding.Traffic congestion uprises traffic hazard incidence, has brought great inconvenience to people's trip.Therefore a kind of vehicle queue length detection method and systematic research based on video is extremely urgent, the queue length of the acquisition vehicle that it can be real-time, and feed back in time the control of intelligent transportation control axis and relieve traffic congestion, improve intelligent level.

Traditional vehicle queue length detection scheme is at porch, track sensor installation formula detecting device.In the publication number patent that is CN102568215A, the Data mining device that adopts toroid winding, coil tuning loop and testing circuit to form, detect flow signal and transmit it to teleseme, by teleseme, the analysis meter of flow signal being calculated the queue length of vehicle.The accuracy of this class sensor type detecting device detection vehicle queue and real-time depend on quantity and the installation site of detecting device largely.In addition on the one hand, once the breaking down of this class sensor type detecting device, maintenance difficulties can be larger.

Existing vehicle queue length detection scheme is the vehicle queue length detection technique based on video image analysis.As the some feature (angle point) by having merged image and line feature (edge), complete wagon flow length detection (referring to Li Yan etc. (12 phases in 2003). application image disposal route detects crossing vehicle queue length automatically. computer utility and software); As the movement of the moving window by artificial demarcation detects vehicle queue length (patent that is CN103258425A referring to publication number).

In the time of actual vehicle queue video analysis, in vehicle queue length detection technique based on video image analysis, the vehicle queue length detection method based on a feature (angle point) and line feature (edge) analysis proposing as above-mentioned journal, due to the problem of the shooting angle of video camera, when vehicle congestion, between vehicle, mutual eclipse phenomena can cause the unstable of the disappearance of a feature and line feature, so that cannot carry out accurately the calculating of follow-up vehicle queue length, finally can not reach the object that detects vehicle queue length.Detect vehicle queue length as above-mentioned patent CN101751558A proposition by the movement of the moving window of artificial demarcation, artificial demarcation moving window described in the method detects and has certain limitation for vehicle queue length, testing result accuracy is had a certain impact, on the other hand, in this technical scheme, only detect for vehicle queue under daytime environment, variation that can not adaptive environment, and the unicity that vehicle characteristics is chosen, by the accuracy that is finally difficult to ensure that vehicle queue length detects.In sum, at present in the urgent need to a kind of method and system that can detect round-the-clock detection vehicle queue length.

Summary of the invention

The object of the embodiment of the present invention is to provide a kind of information real-time based on video camera and the vehicle queue length detection method and the system that build.This system extracts after initial background image from video flowing, sets up camera imaging model and delimits track, finally according to vehicle feature extraction result under different scenes, Vehicle length is adjudicated.This system, in ensureing that necessary information processing real-time is high, considers that round-the-clock environmental change is greatly improved the accuracy detecting.

For meeting above-mentioned purpose, the mist detecting device based on video and the method that provide according to the embodiment of the present invention, the method comprises:

First step, initialization unit, sets up background model by the N two field picture to continuous, obtain stable initialization background image and and initialization reference picture, in initialization reference picture, delimit track, set up camera imaging model;

Second step, by catch a two field picture every time T, calculates this two field picture average brightness value, this value curve over time in adding up one day, analyze the first order derivative of this curve, carry out environment classification according to analysis result, and start the detection system of respective environment.

Third step, based on two kinds of environment classifications: daytime, night.According to the type of environment, extract respectively under current environment clarification of objective and state in area-of-interest, judge whether target is slow running vehicle.

The 4th step, according to the existence of vehicle and motion state, judges whether current area-of-interest to move to next area-of-interest by predefined rule.

The 5th step, circulation third step and the 4th step, until find the tail of the queue of vehicle queue, arrive three-dimensional transformational relation according to two dimensional image, thus the actual queue length of calculating and exporting vehicle.

According to another aspect of the present invention, the vehicle queue length detection system based on video camera that the embodiment of the present invention provides, this system composition structure comprises:

(1) initialization unit, sets up background model by the N two field picture to continuous, obtain stable initialization background image and and initialization reference picture, in initialization reference picture, delimit track, set up camera imaging model;

(2) environment classification unit, by catch a two field picture every time T, calculates this two field picture average brightness value, this value curve over time in adding up one day, analyze the first order derivative of this curve, carry out environment classification according to analysis result, and start the detection system of respective environment.

(3) vehicle identifying unit, based on two kinds of environment classifications: daytime, night.According to the type of environment, extract respectively under current environment clarification of objective and state in area-of-interest, judge whether target is slow running vehicle.

(4) vehicle queue's detecting unit, according to the existence of vehicle and motion state, judges whether current area-of-interest to move to next area-of-interest by predefined rule.

(5) output unit, circulation (3) and (4), until find the tail of the queue of vehicle queue, and according to the transformational relation of 2 d-to-3 d, thus the actual queue length of calculating and exporting vehicle.

Compared with prior art, the key distinction and effect thereof are embodiment of the present invention:

Existing vehicle queue's detection method based on video is all for the detection under daytime environment mostly, detection under night environment is had to less concern, on the other hand, it detects the slip of vehicle queue length based on detection window, but its size to moving window is set and do not done too much introduction, the artificial moving window of setting has larger impact to the accuracy that detects vehicle queue length.In addition, existing detection method, do not consider due to some vehicle reaction not in time, the room phenomenon of the doubtful arrival tail of the queue causing, this can calculate and have a great impact final vehicle queue length.Based on above 3 points, the accuracy that existing vehicle queue length detects adaptive ability to environment and detection needs to improve.

The method that the present invention proposes, in initialization unit, first set up background model by the N two field picture to continuous, obtain stable initialization background image, in initialization background image,, obtain four points in pavement markers with rectangle geometrical property and form rectangle frame in conjunction with manual correction by video camera Corner Detection, complete the foundation of camera model.Further, according to the mapping relations of the 2 d-to-3 d in camera model, taking vehicle length L on actual road surface as according to by driveway partition as several area-of-interests.Compare the size of the artificial setting moving window in existing vehicle queue detection method, accuracy has had raising to a certain degree.In environment classification unit, by catching a two field picture every the T time, calculate the ensemble average brightness of this two field picture, draw ensemble average brightness conditional curve over time.Analyze the first order derivative of this curve, can obtain the trend of environment overall brightness curve, thereby start accurately the detection system of respective environment, improved the environment self-adaption ability that vehicle queue is detected.In vehicle characteristics extraction unit, for automobile storage in daytime environment when the feature extraction, what adopt is that background image and current frame image are carried out respectively to difference, binaryzation and Roborts edge extracting, after obtaining vehicle ' s contour, profile is filled, dilation and erosion, obtain the vehicle image that compactedness is intact, reduce the impact in the vehicle region cavity of causing due to stationary vehicle, thereby improved the accuracy that vehicle characteristics extracts.In vehicle tail of the queue detects, due to some vehicle reaction not in time, the room phenomenon of the doubtful arrival tail of the queue causing, in fact the vehicle of these surveyed areas is still in queue queue, the present invention has increased the processing mode for this phenomenon, can improve to a great extent the accuracy of vehicle queue length.In sum, the method that the present invention proposes has improved the adaptive faculty of detection system to environmental change, has stronger robustness.

With reference to explanation and accompanying drawing hereinafter, and disclosing in detail embodiments of the present invention, other features of the present invention and advantage are set forth, should be appreciated that embodiments of the invention only provide as example, in the spirit of claims and the scope of clause, real-time mode of the present invention comprises many changes, revises and is equal to.

Brief description of the drawings

Fig. 1 is the system chart of the vehicle queue length detection method based on video;

Fig. 2 is the method schematic diagram of initialization unit;

Fig. 3 is the method schematic diagram of environment classification unit;

Fig. 4 is vehicle characteristics and the method schematic diagram of state extraction unit on daytime;

Fig. 5 is vehicle characteristics and the method schematic diagram of state extraction unit at night;

Fig. 6 is the method schematic diagram of vehicle queue length detecting unit;

Fig. 7 is the method schematic diagram that vehicle tail of the queue detects.

Embodiment

Below in conjunction with accompanying drawing, embodiments of the present invention are described in detail, the embodiment of setting forth, only as exemplary illustration, is not limitation of the present invention.For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with accompanying drawing, embodiments of the present invention are described in detail.The specific embodiment of the present invention also needs some common video analysis basic modules, such as image pre-service, context update etc., no longer explains here.The present invention is applicable to freeway toll station and city traffic signal lamp crossing.

Fig. 1 shows according to the system chart of the vehicle queue length detection method based on video of the present invention, mainly comprises 4 parts according to the vehicle queue length detection method based on video of invention:

Part I, initialization unit 101, is used for extracting initial background image and sets up camera model, and the runway region of going forward side by side, selected track to be detected is divided.

Part II, environment classification unit 102, in real time current residing environment being classified, and according to the detection system of classification results startup respective environment.

Part III, vehicle characteristics extraction unit 103, according to the type of environment, extracts the vehicle prominent feature in a certain area-of-interest and the operation characteristic of vehicle under current environment, also judge automobile storage and motion state.

Part IV, vehicle queue's detecting unit 104, according to obtaining vehicle characteristics and state of motion of vehicle in vehicle characteristics extraction unit, controls area-of-interest and moves in selected track, until move to the tail of the queue of vehicle queue, thus calculate vehicle queue length.

Fig. 2 shows according to the method schematic diagram of the initialization unit of the vehicle queue length detection method based on video of the present invention.

In 201, adopt the method for image averaging, obtain main background image by the mean value that calculates N continuous two field picture, formula is as shown in (1).Suppose that present image is I _i, initial background image I _backfor:

$I_{\mathrm{back}}=\frac{1}{N}{\mathrm{\Σ}}_{i=1}^{i=N}{I}_i---\left(1\right)$

Wherein, reference picture is I _ref=I _back.

In 202, by utilizing Harris Corner Detection Algorithm to find out angle point in image, revised by manual, demarcate four some A, B, C, the D with rectangle geometrical property, demarcate track to be detected simultaneously.According to the three's of photo coordinate system, camera coordinate system and world coordinate system transformational relation, set up camera imaging model, by condition AB=CD and AC=BD, obtain the following parameter of video camera: flat-moving angle is that p, pitch angle are that t, rotation angle are the distance l of s, focal distance f and the vertical directed towards ground of video camera minute surface.Thereby set up the mapping relations of two-dimensional image plane (m, n) to three-dimensional world coordinate system (x, y, 0)

, mapping relations formula is as shown in (2) and (3):

$x=\frac{l\sin p(m\sin s+n\cos s)+l\sin t\cos p(m\cos s-n\sin s)}{m\cos t\sin s+n\cos t\cos s+f\sin t}---\left(2\right)$

$y=\frac{-l\cos p(m\sin s+n\cos s)+l\sin t\cos p(m\cos s-n\sin s)}{m\cos t\sin s+n\cos t\cos s+f\sin t}---\left(3\right)$

In 203, choose the starting point (start_x, start_y) of queuing up on lane line to be detected in initialization reference picture and the terminating point (end_x, end_y) of unlimited distance, according to mapping relations

be mapped to the three-dimensional curve that fits to a z=f (x, y) after three-dimensional world coordinate system, wherein z=0.Obtain relational expression as shown in (4):

The corresponding three-dimensional world coordinate (x of the starting point of wherein queuing up (start_x, start_y) ₀, y ₀, 0) put known.Therefore can obtain point set (x equidistant on z=f (x, y) curve _k, y _k, 0), (x _k-1, y _k-1, 0) ... (x ₀, y ₀, 0).Again according to F ^-1, looked like accordingly plane point set (m _k, n _k), (m _k-1, n _k-1) ... (m ₀, n ₀), according to these point sets, will be that several length are the area-of-interest of L as the driveway partition to be detected in plane.Regulation L is 4 meters of the average lengths (comprising the length of vehicle itself and the spacing of its maintenance) of the vehicle on actual road surface.

Fig. 3 shows according to the method schematic diagram of the environment classification unit of the vehicle queue length detection method based on video of the present invention.

In 301, by obtaining a two field picture every the T time, calculate the ensemble average brightness of t time chart picture.If the brightness of certain pixel is Lum (i, j), shown in t moment ensemble average brightness L (t) computing formula (5):

$L\left(t\right)=\frac{1}{{\mathrm{<figure-callout\; id="1"\; label="N"\; filenames="HDA0000494798140000041.png"\; state="\{\{state\}\}">N</figure-callout>}}_1}{\mathrm{\&Sigma;}}_{(i,j)}\mathrm{Lum}(i,j,t)---\left(5\right)$

Wherein N ₁=i*j, represents pixel number, and t represents the time (zero clearing every day).

In 302, according to the Changing Pattern of L (t) value, calculate L ' (t).

In 303, make acting according to 302 result: if L ' is (t) >0, status=1 is set, expression current environment is daytime.

In 304, make acting according to 302 result: if L ' is (t) <0, status=0 is set, expression current environment is night.

Fig. 4 has gone out the vehicle characteristics and the method schematic diagram of state extraction unit on daytime according to the vehicle queue length detection method based on video of the present invention.

In 401, obtain state status=1, start vehicle queue length detection system on daytime.

In 402, to initial background image I _backwith current frame image I _cur, carry out respectively Roborts edge extracting, obtain I ' _backand I ' _curedge image.To I ' _backand I ' _curedge image carries out Difference Calculation and obtains difference image I _chafen, then to I _chafencarry out binaryzation, obtain the I ' of the edge contour information that comprises all vehicles _chafenimage.The threshold value T of regulation binary conversion treatment ₂value is 20.

In 403, to I ' _chafenimage carries out profile filling, obtains the binary image I that comprises complete information of vehicles _result.

In 404, to 403 statistics I _resultthe pixel number N of vehicle in a certain area-of-interest in image ₂, and all pixel number N of this area-of-interest ₃.Calculate N ₂and N ₃ratio p, formula is as shown in (6):

$p=\frac{N_2}{N_3}---\left(6\right)$

If p>T ₂, representing has automobile storage to exist in area-of-interest, otherwise does not have vehicle, wherein specifies T ₂be 0.8.

In 405, obtain three two field pictures, suppose current frame image I _cur, front cross frame image is I _cur-1, I _cur-2, by the poor operation of this three two field pictures frame, formula is as shown in (7) and (8):

I _cha1＝|I _cur-I _cur-1| (7)

I _cha2＝|I _cur-I _cur-2| (8)

To image I _cha1and I _cha2image carries out binaryzation, obtain image I ' _cha1and I ' _cha2, wherein, binary-state threshold parameter is T ₃can get the arbitrary integer between 8～15, T ₄can get the arbitrary integer between 8～15.

In 406, calculate respectively I ' _cha1and I ' _cha2the non-zero pixels points N in the region of vehicle movement in image ₄and N ₅, k the pixel number N that area-of-interest is all _k, obtaining the motion state parameters of vehicle, formula is as shown in (9):

$\mathrm{\&upsi;}=\frac{|N_4-N_5|}{N_k*T_f}---\left(9\right)$

Wherein T _frelevant with frame per second, T in the present invention _fvalue get 0.08.State to υ is adjudicated: if υ≤T ₅, represent that vehicle is in slow running state, otherwise vehicle is in high-speed cruising state, wherein T ₅value get the arbitrary value between 1.5～2.5.

Fig. 5 has gone out the vehicle characteristics and the method schematic diagram of state extraction unit at night according to the vehicle queue length detection method based on video of the present invention.

In 501, obtain state status=0, start vehicle queue length detection system at night.

In 502, current frame image is carried out to white cap transformation, obtain the binary image I that comprises car light information _night.

In 503, at I _nightin image, extract profile boundary information (profile number cnt, the centre coordinate C ' of profile _i(x, y)).Wherein, according to mapping relations F, the profile barycentric coordinates as in plane are mapped to three-dimensional world coordinate C ' _i(x, y), wherein 0<i≤cnt.

In 504, the profile boundary information in a certain area-of-interest is analyzed.Automobile storage condition, formula is as shown in (10) and (11):

cnt＝2 (10)

||C′ ₁(x,y)-C′ ₂(x,y)||＝T ₆ (11)

Meet above two conditions and have vehicle.Wherein T ₆represent the fixed range of vehicle car light, can value 2～2.5 meters.

In 505, obtain three two field pictures, present frame binary image I _nightwith its front cross frame binary image I _night-1, I _night-2, carry out the poor operation of three frames.As shown in formula (12) and (13):

I _ch1＝I _night∩I _night-2 (12)

I _ch2＝I _night∩I _night-1 (13)

In 506, calculate respectively I _ch1and I _ch2the pixel number N in the region of vehicle movement in image ₆and N ₇, and k the pixel number N that area-of-interest is all _k.Obtain the motion state of vehicle, as shown in formula (14):

${\mathrm{\&upsi;}}^{\&prime;}=\frac{{|N}_6-N_7|}{N_k*T_k^{\&prime;}}---\left(14\right)$

Wherein T ' _frelevant with frame per second, T ' in the present invention _fvalue get 0.08.。State to υ ' is adjudicated: if υ '≤T ₇represent that vehicle is in slow running state, otherwise vehicle is in high-speed cruising state.Wherein T ₇get the arbitrary value between 0.3～0.6.

Fig. 6 has gone out according to the method schematic diagram of the vehicle queue length detecting unit of the vehicle queue length detection method based on video of the present invention.

In 601, represent that existing front n-1 car is in queueing condition, the queueing condition of n surveyed area of current detection.

In 602, obtain current detection region automobile storage state, and make judgement: E _n=1, jump to 603; E _n=0, jump to 607.

In 603, obtain the state of current detection region vehicle movement, and make judgement: S _n=1, jump to 604; S _n=0, jump to 607.

In 604, carry out the detection of vehicle tail of the queue, judge whether current detection region is vehicle queue's tail of the queue, and Fig. 7 is shown in by method detailed schematic diagram.

In 605, from 604, obtain the state value IsFind of tail of the queue, and make judgement: IsFing=2, jumps to 606; Otherwise, jump to 608.

In 606, IsFind=2 represents to arrive queuing tail of the queue, and tail of the queue region i.e. n surveyed area.Calculate vehicle queue length.Take two kinds of modes to calculate vehicle queue length, Fuzzy Calculation and accurate Calculation.

Fuzzy Calculation mode, as shown in formula (15):

Length＝n*L； (15)

Accurate Calculation mode:

Obtain the coordinate (x, y) of vehicle queue's tail of the queue, shown in queue length Length computing formula (16):

$\mathrm{Length}=\sqrt{{(x-x_0)}^2+{(y-y_0)}^2}---\left(16\right)$

In 607, control surveyed area and move to n-1 surveyed area.

In 608, control surveyed area and move to n+IsFind+1 surveyed area.

Fig. 7 has gone out the method schematic diagram detecting according to the vehicle tail of the queue of the vehicle queue length detection method based on video of the present invention.

Suppose that in l area-of-interest, automobile storage is E at status flag _l, the motion state of vehicle is S _l, can be defined as follows:

In 701, represent E in n surveyed area _n=1 and S _n=1, in n surveyed area, there is the vehicle of slow running.

In 702, detect n+1 surveyed area E _n+1state, and go out judgement: E _n+1=0, jump to 703; Otherwise, jump to 705.

In 703, detect n+2 surveyed area E _n+2state, and make judgement E _n+2=0, jump to 704; Otherwise, jump to 707.

In 704, represent as IsFind=2 under precondition, represent to find to arrive tail of the queue place of vehicle queue.

In 705, detect n+1 surveyed area S _n+1state, S _n+1=1, jump to 706.Under current situation, based on traffic rules and safe consideration, should there is not S _n+1=0 state.

In 706, represent as IsFind=0 under precondition, represent not arrive tail of the queue place of vehicle queue, surveyed area is moved to n+IsFind+1 surveyed area, proceed to detect.

In 707, detect n+2 surveyed area S _n+2state, and make judgement: S _n+2=1, jump to 708; Otherwise, jump to 709.

In 708, represent that n+1 surveyed area of this state representation do not exist vehicle as IsFind=1 under precondition, and there is not vehicle in n+2 surveyed area, has the vehicle of slow running.Also the vehicle reaction of n+2 surveyed area is not in time, the room phenomenon of the doubtful arrival tail of the queue causing, actually the vehicle of n+2 surveyed area still in queue queue, does not find tail of the queue yet, surveyed area is moved to n+IsFind+1 surveyed area, proceed to detect.

In 709, represent as IsFind=0 under precondition, represent not arrive tail of the queue place of vehicle queue, surveyed area is moved to n+IsFind+1 surveyed area, proceed to detect.

It should be noted that, for the innovation part of outstanding the inventive method, in above-mentioned each equipment embodiment of the present invention, do not mention and realize the relevant conventional elementary cell that this technology is used, but this does not show to implement the elementary cell that the said equipment mode does not need other.

Claims (8)

1. vehicle queue length detection method and the system based on video, in particular, this invents by delimit track in initialisation image, control the movement of surveyed area interested, obtain the status information of surveyed area interested, and convert the queue length of two dimensional image plane to queue length that actual three dimensions is corresponding by setting up camera imaging model, realize the detection of real-time vehicle queue length, described installation composition structure comprises:

(1) initialization unit, sets up background model by the N two field picture to continuous, obtain stable initialization background image and and initialization reference picture, in initialization reference picture, delimit track, set up camera imaging model;

(2) environment classification unit, by catch a two field picture every time T, calculate this value curve over time in this two field picture mean flow rate Data-Statistics one day, analyze the first order derivative of this curve, carry out environment classification according to analysis result, and start the detection system of respective environment;

(3) vehicle identifying unit, based on daytime, night two kinds of environment, according to the type of environment, extracts respectively under current environment clarification of objective and state in area-of-interest, judges whether target is slow running vehicle;

(4) vehicle queue's detecting unit, according to the existence of vehicle and motion state, judges whether current area-of-interest to move to next area-of-interest by predefined rule;

(5) output unit, circulation (3) and (4), until find the tail of the queue of vehicle queue, arrives three-dimensional transformational relation according to two dimensional image, thus the actual queue length of calculating and exporting vehicle.

2. method according to claim 1, it is characterized in that, in initialization reference picture, utilize Harris Corner Detection Algorithm to find out angle point in image, revised by manual, demarcation has four points of rectangle geometrical property, demarcate track to be detected simultaneously, then can obtain the following parameter of video camera by camera calibration model: flat-moving angle is that p, pitch angle are that t, rotation angle are that s, focal length are that f and video camera distance are l, thereby, can obtain two-dimensional image plane to three-dimensional world coordinate system mapping relations , that is:

Wherein, (m, n) is the point on image, and (x, y, 0) is the point of real world coordinate system.

3. method according to claim 2, it is characterized in that, the demarcation in described detection track, specifically, starting point (the start_x of vehicle queue on lane line to be detected will be chosen in initialization reference picture, start_y) and the terminating point (end_x, end_y) of unlimited distance, according to mapping relations

be mapped to the three-dimensional curve that fits to a z=f (x, y) after three-dimensional world coordinate system, according to following relational expression:

The corresponding three-dimensional world coordinate (x of the starting point of wherein queuing up (start_x, start_y) ₀, y ₀, 0) put knownly, can obtain accordingly point set (x equidistant on z=f (x, y) curve _k, y _k, 0), (x _k-1, y _k-1, 0) ... (x ₁, y ₁, 0), then basis

, looked like accordingly plane point set (m _k, n _k), (m _k-1, n _k-1) ... (m ₀, n ₀), according to these point sets, will be k the area-of-interest that physical length is L as the driveway partition to be detected in plane, the average length (comprising the length of vehicle itself and the spacing of its maintenance) of the vehicle that wherein L is actual road surface.

4. method according to claim 1, it is characterized in that, in environment classification unit, catch a two field picture every set time T, the average brightness value of computed image, and draw the average brightness value curve over time of the image that obtains in a day, if curve is in rising trend from plateau, be that slope is greater than zero, be judged as the transition period on night to daytime, switch to detection system on daytime; If curve is on a declining curve from plateau, slope is less than zero, is judged as the transition period at daytime to night, switches to detection system at night.

5. method according to claim 1, is characterized in that, in identifying unit, if testing environment is daytime, can calculate the area of vehicle ' s contour in area-of-interest and the ratio p of area-of-interest area, if p is greater than threshold value T in automobile storage ₁, think and have automobile storage to exist; If testing environment is evening, can detect in area-of-interest whether have car light, if there is car light, think and have automobile storage to exist; In vehicle-state identifying unit, utilize three frames poor after in two width images the difference of vehicle movement pixel number and the ratio υ of time judge the motion state of vehicle.

6. system according to claim 1, is characterized in that, in vehicle queue's detecting unit, according to the existence of vehicle in area-of-interest and motion state, controls area-of-interest and moves, and the movement of described control area-of-interest is according to following principle:

1) if detect that l area-of-interest exists the vehicle of low speed driving, moves to l+1 area-of-interest;

2) if detect that l area-of-interest do not exist vehicle, to l-1 mobile area-of-interest;

3) if detect that l area-of-interest exists the vehicle in fast running, moves to l-1 area-of-interest.

7. system according to claim 1, is characterized in that, in output unit, the tail of the queue judgement of described queuing vehicle can be divided into four kinds of situations (supposing that n area-of-interest exists slow moving vehicle):

1) if detect that n+1 area-of-interest exists vehicle and in low vehicle speeds, do not find temporarily tail of the queue, move to n+1 surveyed area, proceed tail of the queue and detect;

2) if detect that n+1 area-of-interest do not exist vehicle, start to detect n+2 area-of-interest, if n+2 area-of-interest do not exist vehicle, find tail of the queue, tail of the queue region i.e. n surveyed area;

3) if detect that n+1 area-of-interest do not exist vehicle, start to detect n+2 area-of-interest, if there is vehicle in n+2 area-of-interest, and in low speed driving state, tail of the queue is not found in explanation, move to n+2 surveyed area, proceed tail of the queue and detect, this situation is the room phenomenon causing not in time due to vehicle reaction;

4) if detect that n+1 area-of-interest do not exist vehicle, start to detect n+2 area-of-interest, if n+2 area-of-interest exists vehicle, and in high-speed travel state, tail of the queue is not found in explanation, moves to n surveyed area, proceeds tail of the queue and detects.

8. system according to claim 1, is characterized in that, takes two kinds of modes to calculate vehicle queue length, Fuzzy Calculation and accurate Calculation;

Fuzzy Calculation account form: Length=n*L, wherein n is tail of the queue region.

Accurate Calculation mode: obtain the coordinate (x, y) of vehicle queue's tail of the queue, utilize computing formula

obtain queue length.

Images