CN103150900B - Traffic jam event automatic detecting method based on videos - Google Patents

Abstract

The invention discloses a traffic jam event automatic detecting method based on videos. The method includes steps that (1) real-time traffic parameter information of monitoring points is obtained based on a video detecting device, the parameter information is transmitted to a background server to store in a real-time mode; (2) obliterated data are identified, abnormal data are filtered, and data normalization processing is implemented by using a z-score method; (3) historical data are extracted to conduct clustering analysis by using a automatic detecting processing device, road traffic state foundation clustering centers are generated by utilizing a clustering analysis method; (4) euclidean distances to each clustering center are calculated according to the real-time traffic information, a present traffic jam event is automatically judged, and processed data are released through a release terminal device; (5) the clustering centers of each traffic state are recalculated, iteration is repeated, and traffic jam event automatic detection is achieved. The traffic jam event automatic detecting method based on the videos is mainly used for collecting traffic information, reduces cost of traffic information collection and improves veracity and response speed of the traffic jam event automatic detecting.

Description

A kind of traffic congestion event automatic detection method based on video

Technical field

The present invention relates to traffic information collection and traffic behavior perception field, the human cost of traffic information collection can be reduced, improve accuracy and the reaction efficiency of traffic congestion event-state graph.

Background technology

Road traffic running status can describe with different traffic parameters, video detecting device can utilize the image detected to carry out differential variation with image background to compare, obtain the device entering and sail out of information of vehicle target, thus obtain the information such as vehicle passes through, existence, vehicle commander, the speed of a motor vehicle, can the multiple traffic parameter of Real-time Obtaining by cycle statistics protocol analysis, various parameter information returns background server by network real-time Transmission and stores, but in existing traffic events for the using method of video detecting device mostly based on single traffic parameter, as the magnitude of traffic flow, speed, density etc., multiple parametric synthesis is not arranged detection and carry out comprehensive analysis processing, but traffic behavior dynamic change is extremely complicated, be difficult to the characteristic reflecting it by single parameter, single optimum configurations can cause the inaccuracy of differentiation, and be all by manually carrying out video monitoring when using video detecting device at present, waste of manpower and efficiency is lower.

The precision of traffic distinguished number directly affects the traffic flow operation of city road network, directly on current used video detecting device, arranging multiple traffic parameter can cause system to the erroneous judgement of state, the differentiation of mistake may cause road section traffic volume load, formed and stagnate, conventional condition discrimination algorithm, as California algorithm, exponential smoothing, standard deviation etc., traffic behavior is classified by all too stiff setting of the change according to single traffic parameter threshold value, urban road traffic flow amount is extremely complicated, simple being described by means of certain traffic parameter cannot obtain accurate condition discrimination, easy impact detects the degree of accuracy of the magnitude of traffic flow, the process affecting emergency event causes the waste of public administration resource.

Summary of the invention

The object of the present invention is to provide a kind of human cost that can reduce traffic information collection, improve the accuracy of traffic congestion event-state graph and the traffic congestion event automatic detection method based on video of reaction efficiency.

For achieving the above object, present invention employs following technical scheme: a kind of traffic congestion event automatic detection method based on video, the equipment used in the method comprises video detecting device, data communications equipment, for the background server of data storing and filtration, automatic check processing equipment and issue terminal equipment, between described each equipment, signal connects in order, and the method comprises following step:

(1) based on video detecting device, obtain the real-time traffic parameter information of road to be measured chain check point, described traffic parameter comprises flow, occupation rate, car speed, time headway data, and described parameter information returns background server through described data communications equipment real-time Transmission and stores;

(2) obliterated data is identified, and cleaning and filtering is carried out to abnormal data, utilize z-score to carry out the software program of data normalization process;

(3) several detecting devices are chosen, extract each detecting device historical data of month, set up historical data base, data are divided into working day data source and festivals or holidays data source, utilize described automatic check processing equipment to the working days certificate extracted and festivals or holidays data carry out cluster analysis respectively, adopt clustering methodology to calculate and generate that road is very unimpeded, unimpeded, the basic cluster centres of four traffic behaviors of walking or drive slowly and block up;

K-means clustering method model construction based on traffic parameter will carry out computational analysis to all sample datas of often kind of traffic parameter information, draws the cluster of workaday four traffic behavior grades, comprise the following steps:

(1A) the historical data sample extracted in month is divided into 4 initial classes, then using the center of gravity of these 4 classes as initial mean vectors;

(2A) sort out one by one all data samplers except congealing point, adopt Euclidean distance method, each sample is included into congealing point from its that nearest class, such congealing point is updated to the current average of this class, until all samples has all returned class;

(3A) step (2A) is repeated, till all samples all can not be reallocated;

(4) according to real-time transport information, calculate the Euclidean distance with each cluster centre, the shortest cluster centre is chosen in contrast, then automatically judges the traffic congestion of current time road to be measured chain, finally the data of process is released news through issue terminal equipment;

(5) recalculate the cluster centre of all kinds of traffic behavior, wait for that next data calculate, iterate, realize traffic congestion event-state graph.

Beneficial effect of the present invention: the present invention utilizes the video monitoring apparatus on road traffic section to be detected, the vehicle operating traffic parameter information of Real-time Obtaining road chain check point, comprise flow, occupation rate, car speed, time headway data, stored to background data base by Internet Transmission, obliterated data and abnormal data are identified and processes, the historical data extracted one month carries out data normalization process, adopt the disposal route of k-means cluster analysis, generate very unimpeded, unimpeded, four class traffic behavior class center of walking or drive slowly and block up, the data that Real-time Obtaining video detector is passed back, calculate the Euclidean distance with each cluster centre, be judged to be current traffic behavior with the cluster centre of bee-line, generate new cluster centre, wait for the calculating of next data, so iterate, analyze and determine the traffic congestion of described road to be measured chain and unimpeded, the automatic judgement realizing road traffic congestion event detects, make use of traffic flow fully, to flow, occupation rate, car speed, time headway supplemental characteristic carries out overall treatment judgement, improve the accuracy of traffic congestion event-state graph, reduce the time and financial cost of being blocked up by manual video monitoring and controlling traffic, add the speed that emergency event reports and manages, service efficiency and the service level of urban road can be improved.

Accompanying drawing explanation

Fig. 1 is workflow diagram of the present invention;

System equipment connection diagram used in Fig. 2 Fig. 1;

Fig. 3 cluster analysis original sample distribution plan;

Design sketch after Fig. 4 k-means cluster analysis;

Fig. 5 is magnitude of traffic flow schematic diagram in 24 hours;

In figure 1, video detecting device, 2, data communications equipment, 3, background server, 4, automatically check processing equipment, 5, issue terminal equipment

Embodiment

The equipment used in a kind of traffic congestion event automatic detection method workflow based on video as illustrated in fig. 1 and 2 and the method, the equipment used comprises video detecting device 1, data communications equipment 2, for the background server 3 of data storing and filtration, automatic check processing equipment 4 and issue terminal equipment 5, data communications equipment 2 and video detecting device 1 are linked together by cable and are then arranged on the video monitoring erecting frame of road, by data communications equipment 2 and background server 3, between automatic check processing equipment 4 and issue terminal equipment 5, signal connects and carries out Signal transmissions in order, following several treatment step is comprised: (1) is based on the video detecting device 1 in video monitoring equipment based on the method after last equipment connection, the real-time traffic parameter flow of four parameter acquiring road to be measured chain check points is set in video detecting device 1, occupation rate, car speed, time headway data, again parameter information real-time Transmission returned background server 3 and store, (2) identified by background server 3 pairs of obliterated datas, and cleaning and filtering is carried out to abnormal data, enter automatic check processing equipment 4 after filtration, a z-score Standardization Act that can utilize algorithmically compiled is installed in automatic check processing equipment 4 and carries out the software program of data normalization process, (3) extract the historical data of month, road is very unimpeded, unimpeded, the cluster centre of four traffic behaviors of walking or drive slowly and block up to adopt clustering methodology to calculate, (4) according to real-time transport information, the Euclidean distance of calculating and each cluster centre, judges that then the traffic congestion of current time road to be measured chain enters issue terminal equipment 5 and carries out issue automatically openly automatically, (5) recalculate the cluster centre of all kinds of traffic behavior, wait for that next data calculate, again perform step 4, iterate, realize traffic congestion event-state graph.

Differentiation traffic behavior underlying parameter unit as shown in Figure 3 owing to selecting is different, and the order of magnitude is different, do not have a comparability, can not simultaneously overall treatment as judging quota, need to carry out standardization processing to data source, by functional transformation by its data value maps to certain numerical intervals, z-score Standardization Act is the typical way be normalized data, and z-score standardized method carries out standardization based on the average of raw data and standard deviation.

Suppose that sample data integrates as S={ (q ₁, o ₁, v ₁, g ₁), (q ₂, o ₂, v ₂, g ₂) ..., (q _n, o _n, v _n, g _n), wherein (q _i, o _i, v _i, g _i) represent a sample point, q _i, o _i, v _i, g _irepresent flow, occupation rate, speed and time headway respectively, unit be respectively/hour, 1, thousand ms/h and second.

The average of flow $\stackrel{-}{q}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{q}_i/n,$ The standard deviation of flow ${\mathrm{var}}_q=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(q_i-\stackrel{-}{q})}^2};$

The average of occupation rate $\stackrel{-}{o}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{o}_i/n,$ The standard deviation of occupation rate ${\mathrm{var}}_o=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(o_i-\stackrel{-}{o})}^2};$

The average of speed $\stackrel{-}{v}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{v}_i/n,$ The standard deviation of speed ${\mathrm{var}}_v=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(v_i-\stackrel{-}{v})}^2};$

The average of time headway $\stackrel{-}{g}=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{g}_i/n,$ The standard deviation of time headway ${\mathrm{var}}_g=\sqrt{\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{(g_i-\stackrel{-}{g})}^2}.$

Input parameter form after normalization is $(\hat{q}=\frac{q_i-\stackrel{-}{q}}{{\mathrm{vqr}}_q},\hat{o}=\frac{o_i-\stackrel{-}{o}}{{\mathrm{var}}_o},\hat{v}=\frac{v_i-\stackrel{-}{v}}{{\mathrm{var}}_v},\hat{g}=\frac{g_i-\stackrel{-}{g}}{{\mathrm{var}}_g})$

The k-means clustering method of traffic behavior synthetic determination is one of conventional data mining mode, n sample set is divided into k bunch (k<n), sample is allowed to condense to congealing point by certain principle, congealing point is constantly revised or iteration, until each bunch of inner sample standard deviation variance summation reach given threshold value or iteration stable till, finally obtain the center (bunch center can be a virtual point) of each bunch, its objective function is as follows:

$J=\underset{j=1}{\overset{k}{\mathrm{\&Sigma;}}}\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{\left|\right|x_i^{\left(j\right)}-c_j\left|\right|}^2$

Wherein:

The sum of the mono-class sample set of n--;

Bunch number (k<n) that the mono-class sample of k--divides;

The desired value of j--cluster analysis;

--sample point;

C _j--cluster centre;

--the distance metric between sample point and cluster centre;

For several video detectors as separate individuality, each individuality is chosen to the historical data of month, data are divided into two groups: one group to be the data source of regular working day, one group is data sources festivals or holidays, then respectively for setting up corresponding cluster centre matrix on working day and festivals or holidays, the traffic discrimination standard different with festivals or holidays two kinds on working day is formulated.K-means clustering method based on traffic parameter builds model, very unimpeded, unimpeded according to the traffic behavior that will obtain, jogging and congestion level, all sample datas of flow, occupation rate, car speed, time headway often being planted to parameter calculate respectively, comprise three steps:

(1) all historical data samples of month are divided into 4 initial classes, then using the center of gravity (average) of these 4 classes as initial mean vectors;

(2) sort out one by one all data samplers except congealing point, adopt Euclidean distance method, each sample is included into congealing point from its that nearest class, such congealing point is updated to the current average of this class, until all samples has all returned class.

(3) step (2) is repeated, till all samples all can not be reallocated.

First the historical data extracting month makees sample, z-score method is utilized to make standard normalized to it, use k-means cluster analysis, show that, in the cluster of workaday four traffic behavior grades, working day, the center matrix of four traffic behavior cluster centres was as follows:

$M=(\hat{q},\hat{o},\hat{v},\tilde{g})=\begin{array}{cccc}-1.23& 0.78& -0.86& 0.57\\ 0.04& 0.47& -0.18& -0.23\\ -0.79& -0.96& -0.58& 0.36\\ 2.121& -2.69& 3.14& -1.95\end{array}$

Obtained the cluster centre point of each bunch by final mask, four cluster centres are respectively $({\tilde{q}}_2,{\tilde{o}}_2,{\tilde{v}}_2,{\tilde{g}}_2),({\tilde{q}}_3,{\tilde{o}}_3,{\tilde{v}}_3,{\tilde{g}}_3),({\tilde{q}}_4,{\tilde{o}}_4,{\tilde{v}}_4,{\tilde{g}}_4).$

According to real-time transport information, gathering a certain moment obtains one group of real time data (q, o, v, g), obtains through data z-score normalized euclidean distance with each cluster centre is calculated successively to flow, occupation rate, car speed, time headway, each is very unimpeded, unimpeded, jogging and congestion status cluster centre be d1, d2, d3, d4 represent respective Euclidean distance respectively, and cluster centre state corresponding to minor increment is judged to be current time traffic congestion.

D1, d2, d3, d4 computing formula is as follows respectively:

$d1=\sqrt{{(q-{\tilde{q}}_1)}^2+{(o-{\tilde{o}}_1)}^2+{(v-{\tilde{v}}_1)}^2+{(g-{\tilde{g}}_1)}^2)}$

$d2=\sqrt{{(q-{\tilde{q}}_2)}^2+{(o-{\tilde{o}}_2)}^2+{(v-{\tilde{v}}_3)}^2+{(g-{\tilde{g}}_4)}^2)}$

$d3=\sqrt{{(q-{\tilde{q}}_3)}^2+{(o-{\tilde{o}}_3)}^2+{(v-{\tilde{v}}_3)}^2+{(g-{\tilde{g}}_4)}^2)}$

$d4=\sqrt{{(q-{\tilde{q}}_4)}^2+{(o-{\tilde{o}}_4)}^2+{(v-{\tilde{v}}_4)}^2+{(g-{\tilde{g}}_4)}^2)}$

Recalculate the cluster centre of all kinds of traffic behavior, wait for the reception of next real time data, calculate the Euclidean distance of new data, upgrade traffic behavior, again calculate new cluster centre, so iterate, analyze and determine the traffic congestion of described road to be measured chain and unimpeded, the automatic judgement realizing road traffic congestion event detects, and is the design sketch after k-means cluster analysis as shown in Figure 4.

Be built in automatic check processing equipment 4 by above-mentioned algorithm composing software program, working procedure calculates the Euclidean distance of each real-time parameter and cluster centre, and regenerates new cluster centre, carries out data instance checking.The traffic behavior value of 24 hours carries out check analysis, data break 45 minutes from database, to select on September 7,6:00 to 2010 year on the 6th September in 2010 as shown in Figure 5 between 6:00, amounts to 288 traffic behavior values, the time response of traffic behavior.

The present invention make use of traffic flow fully, overall treatment judgement is carried out to flow, occupation rate, car speed, time headway supplemental characteristic, the accuracy of traffic congestion event-state graph can be improved, be decreased through time and financial cost that manual video monitoring and controlling traffic blocks up, increase the speed that emergency event reports and manages, service efficiency and the service level of urban road can be improved.

Claims (1)

1. the traffic congestion event automatic detection method based on video, the equipment used in the method comprises video detecting device, data communications equipment, for the background server of data storing and filtration, automatic check processing equipment and issue terminal equipment, between described each equipment, signal connects in order, it is characterized in that: the method comprises following step:

(1) based on video detecting device, obtain the real-time traffic parameter information of road to be measured chain check point, described traffic parameter comprises flow, occupation rate, car speed, time headway data, and described parameter information returns background server through described data communications equipment real-time Transmission and stores;

(2) obliterated data is identified, and cleaning and filtering is carried out to abnormal data, utilize z-score to carry out the software program of data normalization process;

(3) several detecting devices are chosen, extract each detecting device historical data of month, set up historical data base, data are divided into working day data source and festivals or holidays data source, utilize described automatic check processing equipment to the working days certificate extracted and festivals or holidays data carry out cluster analysis respectively, adopt clustering methodology to calculate and generate that road is very unimpeded, unimpeded, the basic cluster centres of four traffic behaviors of walking or drive slowly and block up;

K-means clustering method model construction based on traffic parameter will carry out computational analysis to all sample datas of often kind of traffic parameter information, draws the cluster of workaday four traffic behavior grades, comprise the following steps:

(1A) the historical data sample extracted in month is divided into 4 initial classes, then using the center of gravity of these 4 classes as initial mean vectors;

(2A) sort out one by one all data samplers except congealing point, adopt Euclidean distance method, each sample is included into congealing point from its that nearest class, such congealing point is updated to the current average of this class, until all samples has all returned class;

(3A) step (2A) is repeated, till all samples all can not be reallocated;

(4) according to real-time transport information, calculate the Euclidean distance with each cluster centre, the shortest cluster centre is chosen in contrast, then automatically judges the traffic congestion of current time road to be measured chain, finally the data of process is released news through issue terminal equipment;

(5) recalculate the cluster centre of all kinds of traffic behavior, wait for that next data calculate, iterate, realize traffic congestion event-state graph.