CN110570658A - Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering - Google Patents
Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering Download PDFInfo
- Publication number
- CN110570658A CN110570658A CN201911012550.4A CN201911012550A CN110570658A CN 110570658 A CN110570658 A CN 110570658A CN 201911012550 A CN201911012550 A CN 201911012550A CN 110570658 A CN110570658 A CN 110570658A
- Authority
- CN
- China
- Prior art keywords
- track
- vehicle
- abnormal
- vehicle track
- tracks
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Traffic Control Systems (AREA)
Abstract
the method for identifying and analyzing the abnormal vehicle track of the intersection based on hierarchical clustering utilizes an LCSS algorithm, a DTW algorithm and the hierarchical clustering algorithm to carry out vehicle track clustering analysis, realizes vehicle track mode division based on intersection canalization information, identifies the abnormal vehicle track of the intersection, further carries out clustering analysis again on the vehicle track in each mode respectively, and finely identifies the abnormal vehicle track of each flow direction, thereby establishing an abnormal track database and providing a support basis for intersection canalization design and information control scheme rationality evaluation. The invention improves the hierarchical clustering effect of the vehicle track, can effectively identify the vehicle track type and the abnormal vehicle track at the intersection, further analyzes the vehicle track in each mode type, and identifies and extracts the abnormal vehicle track in each flow direction, thereby providing effective support for the safety management and the jam management of the traffic conflict at the intersection; by analyzing abnormal tracks in the statistical time period, the problem of intersection canalization organization is effectively identified, and a support basis is provided for the optimization and adjustment of a traffic signal timing scheme.
Description
Technical Field
The invention relates to the field of vehicle track recognition and intersection rationality analysis in the field of traffic control, in particular to an intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering.
Background
With the gathering and growth of motor vehicle data, cities in the country face increasingly severe traffic congestion problems, and therefore, research on vehicle trajectories is necessary for traffic management and dispersion. Until now, many researchers have conducted some research on similarity and abnormal trajectories of vehicle trajectories, such as a similar vehicle trajectory query method [ J ] 2016 (pendunland) based on LCSS, in a corner sword, in poetry, in a field), on the basis of collecting a large amount of vehicle traveling GPS data, forming trajectories based on data cleaning, extracting original trajectory profiles by a Ramer-Douglas-Peucker algorithm, and obtaining similar sub-trajectories based on the LCSS algorithm.
On the other hand, a large number of monitoring cameras such as electric police, bayonets, ball machines and the like are distributed on the city roads at the present stage, so that huge structured data are generated, and the application of the data to realize vehicle track analysis is one of the mainstream of the current research. For example, the CN201710492719.5 of the invention provides a method for identifying the ineffective driving track of a vehicle, which realizes the optimal path recommendation by hierarchical clustering of the identification of the vehicle track and carries out the track path planning aiming at the parking problem; the invention CN201510159009.1 provides an abnormal track detection method based on a wide area distribution traffic system, which determines abnormal traffic track points and abnormal tracks through unsupervised clustering.
the vehicle track analysis research at the present stage mainly focuses on two aspects, namely single vehicle track analysis realized on the basis of position data such as GPS positioning data/mobile phone mobile communication data and the like; the second is to judge the abnormal running track of long-distance vehicles in the whole road network, which is the current shortage of effective classification and judgment of the vehicle track at the road junction and the management and application of the traffic at the road junction of the vehicle track.
Disclosure of Invention
the invention provides a method for identifying and analyzing abnormal vehicle tracks at intersections based on hierarchical clustering, which is characterized by extracting historical video data of monitoring equipment of a road junction dome camera, realizing vehicle track clustering analysis by utilizing an LCSS (Long-term storage service) algorithm, a DTW (dynamic data bus) algorithm and the hierarchical clustering algorithm, dividing vehicle track modes based on intersection canalization information, identifying abnormal vehicle tracks at the intersections, further respectively carrying out clustering analysis again on the vehicle tracks in each mode, finely identifying abnormal vehicle tracks in each flow direction, and analyzing abnormal reasons, thereby establishing an abnormal track database and providing a support basis for intersection canalization design and information control scheme rationality evaluation.
The method for identifying and analyzing the abnormal vehicle track at the intersection based on hierarchical clustering comprises the following steps:
Step 1, collecting intersection canalization information, extracting intersection monitoring video data and finishing data cleaning;
Step 2, analyzing a single track aiming at different vehicle IDs, extracting data characteristic points and determining vehicle tracks;
Step 3, clustering the vehicle tracks at the intersections based on an LCSS algorithm and a hierarchical clustering algorithm, dividing the vehicle track mode types, and identifying normal vehicle tracks and abnormal vehicle tracks;
Step 4, extracting a normal vehicle track in a single mode based on the vehicle track type analyzed in the step 3, further analyzing the vehicle track in the single mode, and identifying an abnormal vehicle track and a normal vehicle track in the single mode;
And 5, analyzing road conditions based on the abnormal vehicle tracks in the steps 3 and 4.
further, the step 1 specifically includes the following sub-steps:
Step 1-1, collecting the type of road junction and channelized information thereof;
Step 1-2, extracting original track points according to different vehicle IDs based on the road monitoring video in a unit time period, and marking the original track points as P (f, x, y), wherein f represents the frame number, and x and y represent the coordinate values of the track points;
Step 1-3, original track points of the vehicle are drawn in two-dimensional coordinates based on the original track points P, important characteristic data and incorrect data are determined to be missing and removed, and original track point cleaning is achieved; the missing important characteristic data is deviation data which can form a short path but deviates from the line trend, and incorrect data can not form a dense scatter set which can not be smoothly connected and can not form a path track.
Further, the step 2 specifically includes the following sub-steps:
Step 2-1, analyzing the original track data P based on the same vehicle ID, and extracting quantitative data from unit time within a statistical time period T to realize track feature point extraction;
step 2-2, drawing tracks based on the extracted feature points, arranging according to the number of f frames, determining vehicle tracks, and recording as TR ═ P | PiI is more than or equal to 1 and less than or equal to n, and n is the number of track points }.
Further, the step 3 specifically includes the following sub-steps:
Step 3-1, controlling a similarity threshold gamma, determining the range of the cluster number K, and determining the optimal cluster number K under the similarity threshold gamma based on a hierarchical clustering algorithmopiand its clustering result;
step 3-2, repeating the step 3-1, adjusting the similarity threshold gamma, and repeatedly solving the optimal clustering cluster number K under different similarity thresholds gammaopiIs marked asWherein l represents a similarity threshold number, wherein the similarity threshold γltaking an integer from the numerical value; at the same time eachMean intra-group distance with optimal cluster numberand group inner distance
Step 3-3 based onSolving different similarity threshold values gammalMean square difference in groupand mean square error between groupsAccording to mean square difference in groupAnd mean square error between groupsDetermining an optimal threshold value gammaopiand the number of optimal groups Kopi；
step 3-4, based on the optimal threshold value gammaopiAnd the optimal cluster number KopiDetermining a clustering result of the vehicle track, identifying a vehicle track mode type, and determining a normal vehicle track and an abnormal vehicle track;
Specifically, the optimal threshold γ is determined based on step 3-3opiAnd the optimal cluster number KopiAnd (3) carrying out vehicle track hierarchical clustering, classifying the clustered vehicle track modes according to the intersection canalization information in the step (1), defaulting the vehicle tracks except for downward division according to the intersection canalization flow direction as abnormal vehicle tracks, and taking the rest as normal vehicle tracks.
Further, the step 3-1 comprises the following sub-steps:
Step 3-1-1, controlling a similarity threshold gamma, and determining the longest common subsequence and the longest common subsequence similarity distance between every two tracks through an LCSS algorithm;
Step 3-1-2, listing a similarity matrix S [ a ] [ b ] according to the longest common subsequence similarity distance in the previous step, namely an adjacent matrix;
Step 3-1-3, determining the range of the cluster number K; specifically, determining the minimum value of the clustering cluster number K according to the intersection channelized information in the step 1;
step 3-1-4, giving a cluster number K, and continuously repeating the step 3-1-1 and the step 3-1-2 to obtain a hierarchical clustering result;
Step 3-1-5, determining clusters with different control similarity threshold gamma-timing K values according to the cluster range given in the step 3-1-3under the condition, an evaluation system is established according to the intra-group distance and the inter-group distance under different K value clustering, so that the optimal clustering cluster number K is determinedopiThe method comprises the following steps:
Extracting all characteristic points P of the tracks in the group according to the clustering result of the step 3-1-4iAccording to the characteristic point PiDetermining K cluster centers { C ] under the condition of determining K groups of partitions1，C2，C3，...，CkK is more than 1 and less than or equal to K, wherein the center C of the clusterkis the center of all feature points in the group;
Respectively solving the group inner distance tau in k groups according to the k group cluster centerskAnd inter-group spacing
Group internal distance mean value based on different group numbers KAnd distance between groupsdetermining an optimal number of clusters KopiSpecifically, the ratio isThe maximum K value defaults to Kopi。
Further, the step 3-3 comprises the following sub-steps:
step 3-3-1, according toSolving different similarity threshold values gammalMean square difference in group
Step 3-3-2, according toSolving different similarity threshold values gammalMean square between groupsdifference (D)
Step 3-3-3, based on different similarity threshold values gamma and K thereofopiMean square difference in groupAnd mean square error between groupsdetermining an optimal threshold value gammaopi. Specifically, the ratio isGamma at maximum timelDefaults to an optimal threshold.
Further, the step 4 specifically includes the following sub-steps:
step 4-1, extracting a normal vehicle track in a single mode, and based on the track characteristic point numerical value of the normal vehicle track, using the track similarity lambda and the acceleration variance alpha of the track2taking the arc length ratio sigma as a characteristic numerical value to perform hierarchical clustering, and dividing an abnormal driving track, an abnormal behavior track and a normal vehicle track;
step 4-2, integrating the normal vehicle tracks obtained by analysis in the previous step, performing hierarchical clustering on the normal vehicle tracks in the previous step again under a DTW (dynamic time warping) algorithm, and judging normal vehicle tracks and outlier tracks;
And 4-3, repeating the steps 4-1 to 4-2, analyzing the vehicle track in each mode of the road junction, and dividing the abnormal driving track, the abnormal behavior track, the outlier track and the normal vehicle track in each flow direction, wherein the abnormal driving track, the abnormal behavior track and the outlier track are all defaulted as the abnormal vehicle track.
further, the step 4-1 specifically comprises the following sub-steps:
step 4-1-1, vehicle track TRjTrack similarity lambda and acceleration variance alpha2Solving the arc length ratio sigma;
Determining a similarity calculation formula, such as an LCSS algorithm or a DTW algorithm, according to the vehicle track characteristics to obtain a vehicle track similarity lambda;
Determining acceleration variance alpha of each vehicle track based on track characteristic points of vehicle tracks TR2namely:
In the formula, alphai+1represents the acceleration of the characteristic point i +1, where fiand fi+1Which represents the number of frames,denotes the Euclidean distance, p, between the feature point i and the feature point i +1iAnd pi+1Representing the feature points;
Determining a vehicle trajectory arc length ratio σ based on the vehicle trajectory TR, i.e.:
in the formula, p1、pi、pi+1、pneach represents a characteristic point within the vehicle trajectory;Representing the Euclidean distance between the vehicle track characteristic point i and the characteristic point i + 1;Representing the Euclidean distance between the vehicle track characteristic point n and the characteristic point 1;
Step 4-1-2, similarity lambda and acceleration variance alpha based on the vehicle track TR in the previous step2Taking the arc length ratio sigma value as characteristic data to perform hierarchical clustering, and dividing the characteristic data into an abnormal driving track, an abnormal behavior track and a normal vehicle track; specifically, the degree of similarity λ and the acceleration variance α are used2taking the arc length ratio sigma value as characteristic data to perform hierarchical clustering, dividing the vehicle track into three groups of data, and determining the data type according to the data quantity of the three groups of data and the discrete degree of each group of data;
Calculating the track number of the three groups of vehicle tracks and the ratio of the track number to the total track number, and if the track number in the group is the least, the data of the group of vehicle tracks is defaulted as an abnormal running track;
Respectively solving the dispersion degree epsilon of the rest two groups of vehicle track data based on the characteristic points of the vehicle tracks, namely:
In the formula, N is the total number of all vehicle track characteristic points in the group; p is a radical ofiIn order to be a characteristic point, the method comprises the following steps of,Is the cluster center point;represents piDistance from point to cluster center point;
And defaulting the group of data with larger dispersion degree epsilon as an abnormal behavior track, defaulting the group of data with smaller dispersion degree epsilon as a normal vehicle track, and simultaneously realizing normal vehicle track extraction.
Further, the step 5 specifically includes the following sub-steps:
Step 5-1, integrating the abnormal vehicle tracks in the steps 3 and 4, analyzing the reasons for the abnormal vehicle tracks, meanwhile, corresponding the abnormal vehicle tracks to the abnormal reasons, and establishing an abnormal track association database;
And 5-2, analyzing the road junction canalization organization problem and the traffic signal control problem based on the abnormal track number proportion and the behavior reason thereof in the statistical time period, and identifying unreasonable road junction canalization and unreasonable road junction signal schemes.
further, in the step 5-2, the following is specifically performed:
if the number of the abnormal tracks in the statistical time period is larger than the threshold value of the abnormal condition of the intersection, analyzing whether canalization has a problem or not;
If the same flow is downward, counting the number of abnormal tracks in the time period release phase stage, and if the number of abnormal tracks is greater than the threshold value of the abnormal condition, analyzing whether the signal scheme configuration of the phase stage is reasonable.
The invention achieves the following beneficial effects:
1. The invention relates to a method for analyzing vehicle track similarity by using a video number plate, which comprises the steps of relying on GPS data and road network information, matching GPS point location information with a road network map to realize data preliminary processing and cleaning, realizing clustering through GPS data characteristic points, replacing Euclidean distance with LCSS similarity length distance to realize hierarchical clustering, but not judging the clustering effect.
2. Compared with the traditional road network vehicle track research (integrated cluster analysis of all vehicle tracks), the method analyzes a single intersection and a single flow direction vehicle track, classifies the intersection vehicle track through an LCSS hierarchical clustering algorithm, effectively identifies the intersection vehicle track type and the abnormal track, further analyzes the vehicle track in each mode type, and identifies and extracts the abnormal track of each flow direction vehicle, thereby providing effective support for intersection traffic conflict safety management and congestion management.
3. The invention creatively establishes an abnormal track database (abnormal vehicle track types and track abnormal reasons corresponding to one), analyzes abnormal vehicle tracks according to intersection canalization information and a traffic signal scheme, effectively identifies the intersection canalization organization problem and the signal scheme problem by analyzing the abnormal tracks in a statistical time period, and provides a support basis for optimizing intersection canalization and optimizing and adjusting the traffic signal timing scheme.
Drawings
Fig. 1 is a flowchart illustrating steps of a vehicle trajectory recognition and analysis method according to the present invention.
Fig. 2 is a schematic diagram illustrating an original track point of a vehicle according to an embodiment of the present invention.
FIG. 3 is a schematic diagram of a screened vehicle trajectory according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a vehicle track labeled with colors according to clustering information in the embodiment of the present invention.
fig. 5 is a schematic diagram of an abnormal vehicle trajectory extracted in the embodiment of the present invention.
fig. 6 is a list of similarity, arc length ratio and acceleration variance of each vehicle trajectory obtained by solving the left-turn vehicle trajectory of the south entry lane in the embodiment of the present invention.
fig. 7 is a schematic diagram of three vehicle trajectories in which the vehicle trajectory is divided into a normal trajectory, an abnormal behavior, and an abnormal trajectory by hierarchical clustering in the embodiment of the present invention.
Fig. 8 is a schematic diagram of a divided normal track and an outlier track in the embodiment of the present invention.
FIG. 9 is a schematic diagram of all vehicle trajectories in an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further explained in detail by combining the drawings in the specification.
The method for identifying and analyzing the abnormal vehicle track of the intersection based on hierarchical clustering can analyze the video structured data, realize the identification and classification of the track types of the intersection, effectively mark off the abnormal vehicle track, establish a track database for the abnormal track, and assist in realizing the traffic canalization and the evaluation and analysis of a signal control scheme. The method specifically comprises the following steps:
Step 1, intersection channelizing information is collected, intersection monitoring video data is extracted, and data cleaning is finished.
step 1-1, the type of the road junction and the channelizing information thereof are collected. If the crossroad is adopted, channelized information acquisition is carried out on the number and the width of lanes of the east, south, west and north entrance lanes and the width of lanes of the non-motor vehicles; and if the overpass intersection exists, acquiring lane information of the turnout intersection.
and 1-2, extracting original track points according to different vehicle IDs based on the road monitoring video in a unit time period, and recording the original track points as P (f, x, y), wherein f represents the frame number, and x and y represent the coordinate values of the track points. Generally, a monitoring video comes from an electronic police, an intelligent gate, a dome camera monitoring and the like, and mainly comprises a dome camera video with a high angle and a good visual field.
And 1-3, drawing original track points of the vehicle in a two-dimensional coordinate based on the original track points P, determining missing important characteristic data and incorrect data and removing the missing important characteristic data and the incorrect data to realize cleaning of the original track points. The missing important characteristic data is deviation data which can form a short path but deviates from the line trend, and incorrect data can not form a dense scatter set which can not be smoothly connected and can not form a path track.
And 2, analyzing the single track aiming at different vehicle IDs, extracting the data characteristic points of the single track and determining the vehicle track.
and 2-1, analyzing based on original track data P under the same vehicle ID, and selecting n data from each unit time T in a statistical time period T to realize track characteristic point extraction. Generally, the unit time t is selected to be 5-6 seconds, n is selected to be 25-35, and meanwhile, for convenience of clustering analysis and improvement of clustering effect, the total characteristic point of each vehicle track can be a numerical value.
Step 2-2, drawing a track based on the extracted feature points P, arranging according to the number of f frames, determining a vehicle track, and recording as TR ═ P | Pii is more than or equal to 1 and less than or equal to n, and n is the number of track points }.
and 3, realizing intersection vehicle track clustering based on an LCSS algorithm and a hierarchical clustering algorithm, dividing a vehicle track mode type, and identifying a normal vehicle track and an abnormal vehicle track.
step 3-1, controlling a similarity threshold gamma, determining the range of the cluster number K, and determining the optimal cluster number K under the similarity threshold gamma based on a hierarchical clustering algorithmopiAnd its clustering result.
Step 3-1-1, controlling a similarity threshold gamma, and determining the longest common subsequence and the longest common subsequence similarity distance between every two tracks through an LCSS algorithm, namely:
Wherein TR is1and TR2Two tracks with the length of m and n are respectively arranged;Wherein p isiand q isjRespectively represent the coordinates of the characteristic points of the track,In the same way Wherein i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to n,indicating that the track is empty; dist (p)i,qj) Representing Euclidean distance of two coordinate points, wherein gamma is a similarity threshold value; LCSS (TR)1,TR2) Represents TR1And TR2The longest common subsequence length of the two tracks; dLCSS(TR1,TR2) Is a track TR1And TR2the similarity distance of (2); min (len)TR1,lenTR2) Representing the track TR1Length and TR2The smaller value of the length.
And 3-1-2, listing a similarity matrix S [ a ] [ b ] according to the longest common subsequence similarity distance in the previous step, namely the adjacent matrix.
Specifically, the Euclidean distance in the traditional hierarchical clustering is replaced by the LCSS similarity distance, and the similarity distances between every two tracks are listed respectively, wherein the i-th row example is used, the value in the row is the similarity distance of the longest public subsequence, namely { DLCSS(TRj,TR1),DLCSS(TRj,TR2),…,DLCSS(TRj,TRN) J is more than or equal to 1 and less than or equal to N in the formula, and represents the ith vehicle track.
And 3-1-3, determining the range (minimum value) of the cluster number K.
specifically, the minimum value of the cluster number K is determined according to the intersection channelized information in the step 1. For example, at an intersection, there are left turn, straight turn and right turn in the east, south, west and north directions, and the intersection is allowed to turn around, so the cluster number K is taken from 16.
And 3-1-4, giving the cluster number K, and continuously repeating the steps 3-1-1 and 3-1-2 to obtain a hierarchical clustering result. The method specifically comprises the following steps:
1) And continuously merging the two closest tracks according to the similarity value of the longest common subsequence, specifically, taking min (SA) (b) for merging, wherein the track coincidence condition is eliminated (the value is zero).
Such as TR1、TR2、TR3、TR4Four vehicle tracks, and the longest common subsequence similarity distance, the proximity matrix S [ a ] is solved][b]Comprises the following steps:
then TR can be adjusted1And TR4Merge (value 0.6) as a new clusterAnd is subjected to S [ a ]][b]And solving and clustering.
2) Defaulting the two merged tracks into a new cluster, and repeating the steps until the set number of K clusters is reached. Specifically, TRjAnd TRj-1d is recalculated when the second round of hierarchical clustering is carried out according to the merged clustering result of the previous roundLCSSnumerical values and new S [ a ] are listed][b]And (4) matrix.
Step 3-1-5, determining clustering conditions with different control similarity threshold gamma and timing K values according to the cluster range given in the step 3-1-3, and establishing an evaluation system according to the intra-group distances and inter-group distances under the clustering of different K values, thereby determining the optimal clustering cluster number Kopi. The method specifically comprises the following steps:
1) Extracting all characteristic points P of the tracks in the group according to the clustering resultiAccording to the characteristic point PiDetermining K cluster centers { C ] under the condition of determining K groups of partitions1，C2，C3，...，CkK is more than 1 and less than or equal to K, wherein the center C of the clusterkIs the center of all feature points in the set.
2) Respectively solving the group inner distance tau in k groups according to the k group cluster centerskNamely:
In the formula: n is the number of samples in the kth group, wherein K is more than or equal to 1 and less than or equal to K; dist (P)i,Ck) Representing the distance from the sample point in the k groups to the cluster center point; piThe sample information of the track i is more than or equal to 1 and less than or equal to n; ckthe cluster center point for the k groups.
Further, the mean value of the distances between groups under the condition of K groups is obtained and recorded asNamely:
3) solving group spacing based on cluster center point
In the formula: k is the number of cluster centers, CkAnd Ck′Is the cluster center point coordinate.
4) Group internal distance mean value based on different group numbers KAnd distance between groupsDetermining an optimal number of clusters Kopi. Specifically, the ratio isThe maximum K value defaults to Kopi。
Step 3-2, repeating the step 3-1, adjusting the similarity threshold gamma, and repeatedly solving the optimal clustering cluster number K under different similarity thresholds gammaopiIs marked asWherein gamma islRepresenting a similarity threshold, the value being an integer; at the same time eachmean intra-group distance with optimal cluster numberAnd group inner distance
step 3-3 based onSolving different similarity threshold values gammalmean square difference in groupAnd mean square error between groupsDetermining an optimal threshold value gammaopiAnd the number of optimal groups Kopi。
Step 3-3-1, according toSolving different similarity threshold values gammalMean square difference in groupNamely:
In the formula:andThe similarity threshold is gammaltime, optimal cluster number Kopisolving formula of the group internal distance and the group internal mean value of each K groups is S45, wherein K is the cluster number, and K is more than 1 and less than or equal to KopiAnd l is a similarity threshold sequence number.
Step 3-3-2, according tosolving different similarity threshold values gammalMean square error between groupsNamely:
In the formula:Expressed as a similarity threshold of gammaltime, optimal cluster number KopiThe distance between the centers of each k sets of clusters,Expressed as the threshold of the l-th similarity, the optimal clustering number KopiTime group inner distance values; wherein K is the number of clusters, and K is more than 1 and less than or equal to KopiAnd l is a similarity threshold.
step 3-3-3, based on different similarity threshold values gamma and K thereofopimean square difference in groupAnd mean square error between groupsDetermining an optimal threshold value gammaopi. Specifically, the ratio isthe maximum value of l defaults to the optimal threshold.
Step 3-4, based on the optimal threshold value gammaopiAnd the optimal cluster number Kopiand determining a clustering result of the vehicle track, identifying a vehicle track mode type, and determining a normal vehicle track and an abnormal vehicle track.
Specifically, the optimal threshold γ is determined based on step 3-3opiAnd the optimal cluster number Kopiclustering vehicle tracks, classifying the clustered vehicle track modes according to the intersection canalization information in the step 1, and downwards dividing vehicles except vehicles according to intersection canalization streamsThe default of the vehicle track is an abnormal vehicle track, and the rest are normal vehicle tracks, and under the general condition, the abnormal vehicle track at the intersection comprises abnormal conditions such as a non-motor vehicle running track, a vehicle reverse running track, vehicle pause and the like.
And 4, extracting a single-mode (flow direction) normal vehicle track based on the vehicle track type analyzed in the step 3, further analyzing the single-mode (flow rate) vehicle track, and identifying an abnormal track and a normal vehicle track.
Step 4-1, extracting a normal vehicle track with a single mode (flow direction), and based on the characteristic point value of the vehicle track TR, using the track similarity lambda and the acceleration variance alpha of the track2and the arc length ratio sigma is used as a characteristic numerical value to carry out hierarchical clustering, and an abnormal driving track, an abnormal behavior track and a normal vehicle track are divided.
step 4-1-1, vehicle track TRjtrack similarity lambda and acceleration variance alpha2And the arc length ratio sigma is solved. The method specifically comprises the following steps:
1) And determining a similarity calculation formula, such as an LCSS algorithm or a DTW algorithm, according to the vehicle track characteristics, and further obtaining the vehicle track similarity lambda.
if the LCSS algorithm is adopted, the calculation formula for obtaining the vehicle track similarity lambda based on the LCSS algorithm is as follows:
Wherein TR is1and TR2Two tracks with the length of m and n are respectively arranged;Wherein p isiAnd q isjRespectively represent the coordinates of the characteristic points of the track,In the same way Wherein i is more than or equal to 1 and less than or equal to m, j is more than or equal to 1 and less than or equal to n,indicating that the track is empty; dist (p)i,qj) Representing Euclidean distance of two coordinate points, wherein gamma is a similarity threshold value; LCSS (TR)1,TR2) Represents TR1And TR2the longest common subsequence length of the two tracks; dLCSS(TR1,TR2) Is a track TR1and TR2The similarity distance of (2); min (len)TR1,lenTR2) Representing the track TR1length and TR2The smaller value of the length; dLCSS(TR1,TRs) Represents TR1and TRsThe longest common subsequence similarity distance between the two, wherein s is more than or equal to 1 and less than or equal to N, and N represents the number of vehicle tracks; lambda (TR)1) Represents TR1Similarity values of vehicle trajectories.
If the DTW similarity is adopted, the calculation formula for obtaining the vehicle track similarity lambda based on the DTW algorithm is as follows:
DTW(TR1,TR2)＝f(m,n)
wherein TR1 and TR2 are two respectivelya track of length m, n; TR (transmitter-receiver)1＝[p1,p2,…,pm-1,pm]
TR2＝[q1,q2,…,qn-1,qn]Wherein p isiAnd q isjRespectively represent the coordinates of the characteristic points of the track,In the same way wherein i is more than or equal to 1 and less than or equal to m, and j is more than or equal to 1 and less than or equal to n; i pi-qj| | represents the euclidean distance of two coordinate points; lambda (TR)1) Represents TR1and the similarity value of the vehicle tracks, wherein s is more than or equal to 1 and less than or equal to N, and N represents the number of the vehicle tracks.
Generally, the choice of the LCSS algorithm and DTW algorithm is determined by the intersection characteristics and the trajectory data volume.
2) Determining acceleration variance alpha of each vehicle track based on track characteristic points of vehicle tracks TR2Namely:
In the formula, alphai+1Represents the acceleration of the characteristic point i +1, where fiAnd fi+1Which represents the number of frames,Denotes the Euclidean distance, p, between the feature point i and the feature point i +1iAnd pi+1To representAnd (4) feature points.
3) Determining the vehicle trajectory arc length ratio sigma based on the vehicle trajectory TR, i.e.
in the formula, p1、pi、pi+1、pnEach represents a characteristic point within the vehicle trajectory;representing the Euclidean distance between the vehicle track characteristic point i and the characteristic point i + 1;And representing the Euclidean distance between the vehicle track characteristic point n and the characteristic point 1.
step 4-1-2, similarity lambda and acceleration variance alpha based on the vehicle track TR in the previous step2And the arc length ratio sigma value is used as characteristic data to carry out hierarchical clustering (the clustering group number K is 3), and the characteristic data is divided into abnormal vehicle tracks, abnormal behavior tracks and normal vehicle tracks.
Specifically, the degree of similarity λ and the acceleration variance α are used2And performing hierarchical clustering by taking the arc length ratio sigma value as characteristic data, dividing the vehicle track into three groups of data, and determining the data type according to the data volume of the three groups of data and the discrete degree of each group of data. The method comprises the following specific steps:
1) and calculating the track number of the three groups of vehicle tracks and the ratio of the track number to the total track number, and if the track number in the group is the least, the group of vehicle track data is defaulted to be an abnormal running track, and turning to the next step.
2) Respectively solving the dispersion degree epsilon of the rest two groups of vehicle track data based on the characteristic points of the vehicle track, namely
In the formula, N is the total number of all vehicle track characteristic points in the group; p is a radical ofiin order to be a characteristic point, the method comprises the following steps of,Is the cluster center point;represents pipoint to cluster center point distance.
And defaulting the group of data with larger dispersion degree epsilon as an abnormal behavior track, defaulting the group of data with smaller dispersion degree epsilon as a normal vehicle track, and simultaneously realizing normal vehicle track extraction.
And 4-2, integrating the normal vehicle track of the previous step, performing hierarchical clustering on the vehicle track again under the DTW (the number of clustered cluster groups is 2) and judging the normal vehicle track and the outlier track. The method comprises the following specific steps:
1) Extracting a normal vehicle track, and determining the similarity between every two tracks based on a DTW algorithm, namely:
DTW(TR1,TR2)＝f(m,n)
Wherein TR1 and TR2 are two tracks with length m and n respectively; TR (transmitter-receiver)1＝[p1,p2,…,pm-1,pm]
TR2＝[q1,q2,…,qn-1,qn]Wherein p isiAnd q isjRespectively represent the coordinates of the characteristic points of the track,In the same way Wherein i is more than or equal to 1 and less than or equal to m, and j is more than or equal to 1 and less than or equal to n; i pi-qjand | | represents the euclidean distance of two coordinate points.
Further determining similarity matrix SA between two tracks][b]In the i-th row example, the intra-row value is DTW track similarity, i.e., { D }LCSS(TRj,TR1),DLCSS(TRj,TR2),…,DLCSS(TRj,TRN) J is more than or equal to 1 and less than or equal to N in the formula, and represents the jth track; n denotes the total number of extracted normal vehicle trajectories.
2) And (5) using the similarity to prove each similarity value in the S [ a ] [ b ], and dividing the normal vehicle track into two types by a hierarchical clustering algorithm.
3) for the two types of vehicle track analysis, if the ratio of the number of the certain group of vehicle tracks to the total number of the vehicle tracks is smaller than an abnormal threshold (generally 30%), the combined data is defaulted to be an outlier track, the other combined data is a normal vehicle track, and otherwise, the next step is carried out.
4) And solving two combined vehicle track discrete degrees epsilon based on the characteristic points of each vehicle track, namely:
In the formula, N is the total number of all vehicle track characteristic points in the group; p is a radical ofiin order to be a characteristic point, the method comprises the following steps of,Is the cluster center point;Represents a characteristic point piDistance to cluster center point. The group of data with a large degree of dispersion epsilon is defaulted as an outlier trajectory, and the other group is a normal vehicle trajectory.
And 4-3, repeating the steps 4-1 to 4-2, analyzing the vehicle track in each mode (flow direction) of the intersection, and dividing the abnormal driving track, the abnormal behavior track, the outlier track and the normal vehicle track in each flow direction, wherein the abnormal driving track, the abnormal behavior track and the outlier track are all defaulted to be the abnormal vehicle track.
and 5, analyzing road conditions based on the abnormal vehicle track in the vehicle track identification information in the steps 3 and 4.
And 5-1, integrating the abnormal vehicle tracks in the steps 3 and 4, analyzing the reasons of the abnormal tracks (such as illegal vehicle behaviors, pedestrian-avoiding abnormal behaviors and vehicle conflict behaviors), and simultaneously corresponding the abnormal vehicle tracks to the abnormal reasons to establish an abnormal track association database.
And 5-2, analyzing the road junction canalization organization problem and the traffic signal control problem based on the abnormal track number proportion and the behavior reason thereof in the statistical time period, and identifying unreasonable road junction canalization and unreasonable road junction signal schemes.
1) if the number of the abnormal tracks of a certain reason in the statistical time period is larger than the threshold value of the abnormal condition of the intersection (generally, 30-40 percent is taken, and the abnormal tracks are determined according to the type of the intersection canalization), whether the intersection canalization problem has a problem or not is analyzed according to the reason. And if the vehicles are all subjected to the abnormal behaviors of avoiding pedestrians at a certain intersection within the statistical time period (7 days), judging whether the canalization between the non-motor vehicle lane and the motor vehicle lane is reasonable.
2) If the same flow is downward, counting that the number of abnormal tracks is greater than the threshold value (generally between 30% and 40%) of the abnormal conditions of the intersection in the release phase stage of the time period, analyzing whether the intersection signal scheme configuration is reasonable according to reasons, and if a left-turn phase stage has more abnormal track numbers, whether the left-turn phase sequence or the green light time length setting is reasonable.
The following describes a vehicle track recognition and analysis method according to the present invention by using a specific example.
example 1 is an intersection, a dome camera monitoring video within 5 minutes of a certain intersection is selected, vehicle ID, frame number, X coordinate and Y coordinate values are extracted from the video, an original track point of a vehicle is drawn based on the X/Y coordinate, incorrect data are removed from the original track point, and important characteristic data are deleted, as shown in fig. 2.
Through step 3, feature points of the vehicle are screened based on the ID of the vehicle, for example, original data points under the ID are extracted, 30 pieces of data are randomly screened in 6s unit time of the original data points, and a vehicle track is drawn, as shown in fig. 3 specifically.
And then eliminating the data of the conditions (data points do not move) such as waiting for red lights and the like, screening 100 characteristic points from the total time of 5min, and extracting the characteristic points of the tracks under all the vehicle IDs (305 vehicle tracks in total) in the same way.
Solving the similarity relation among 305 tracks based on LCSS algorithm, wherein the initial value of the similarity threshold value gamma is 30, and listing a similarity matrix S [ a ]][b]will TR1、TR2、TR3、TR4Four data are carried out independently, and the similarity matrix is as follows:
Namely, the similarity between the trajectory TR4 and the trajectory TR1 is 0.6.
And further continuously superposing, calculating and determining the clustering effect under the K groups, comparing the characteristic point inter-group distance numerical values and the group intra-group distance average values of clustering under each K groups, and setting the K value with the maximum ratio as the optimal cluster number, namely determining the cluster number K to 44 under the condition that gamma is 30. And further adjusting the similarity threshold gamma to obtain different similarity thresholds and cluster numbers thereof.
Finally, the optimal threshold value gamma is determined 26 through the group distance mean square error and the group distance mean square error, and the cluster number K is determined 42. On the basis of the feature point trajectory profile, labeling with colors according to clustering information specifically includes left turn (east left turn, west left turn, north left turn, south left turn), straight run (east straight run, west straight run, north straight run, south straight run), right turn (east right turn, west right turn, north right turn, south right turn), u-turn (south u-turn, north u-turn) and other abnormal trajectories, specifically as shown in fig. 4.
The intersection classification situation is shown in fig. 5, and an abnormal vehicle track is extracted from the intersection classification situation.
further, the left-turn vehicle track of the south entry lane is extracted and analyzed, the similarity of each vehicle track is solved based on an LCSS algorithm, the arc length ratio and the acceleration variance of each track are solved at the same time, and a data list is shown in FIG. 6.
The vehicle tracks are divided into normal vehicle tracks, abnormal behavior tracks and abnormal driving tracks through hierarchical clustering, and the specific classification is shown in fig. 7.
Further extracting and analyzing normal vehicle tracks, and solving a similarity matrix between the tracks by using a DTW algorithm, wherein the similarity matrix is as follows:
Thereby dividing a normal vehicle trajectory and an outlier trajectory, as shown in fig. 8.
And analyzing the vehicle tracks in each flow direction, integrating all abnormal vehicle tracks, and finding the reasons of the abnormal tracks, which are mainly vehicle illegal behaviors (detour driving), by comparing the monitored videos when the numerical values of the abnormal vehicle tracks do not exceed the threshold value of the abnormal conditions of the intersection (the number of the abnormal vehicles is less than 35 percent of the total number of the tracks).
example 2 is an overpass intersection, a dome camera monitoring video within 5 minutes of a certain overpass intersection is selected, vehicle ID, frame number, X coordinate and Y coordinate values are extracted from the overpass intersection, original track points of a vehicle are drawn based on the X/Y coordinates, incorrect data are removed from the original track points, important characteristic data are deleted, and original track cleaning and track characteristic point extraction are achieved.
And further continuously iterating through an LCSS algorithm and hierarchical clustering to find an optimal similarity threshold value gamma and the cluster number K (gamma is 30, and K is 10), drawing vehicle track information, determining the vehicle track type, and analyzing abnormal vehicles, wherein the specific categories comprise left turn (in and out), straight travel (south to north and north to south), right turn (in and out) and other abnormal tracks. All vehicle trajectories are shown in fig. 9.
And further analyzing the vehicle track in a single mode, and finally integrating all abnormal tracks to provide a support basis for traffic management.
the above description is only a preferred embodiment of the present invention, and the scope of the present invention is not limited to the above embodiment, but equivalent modifications or changes made by those skilled in the art according to the present disclosure should be included in the scope of the present invention as set forth in the appended claims.
Claims (10)
1. The method for identifying and analyzing the abnormal vehicle track at the intersection based on hierarchical clustering is characterized by comprising the following steps: the method comprises the following steps:
Step 1, collecting intersection canalization information, extracting intersection monitoring video data and finishing data cleaning;
Step 2, analyzing a single track aiming at different vehicle IDs, extracting data characteristic points and determining vehicle tracks;
step 3, clustering the vehicle tracks at the intersections based on an LCSS algorithm and a hierarchical clustering algorithm, dividing the vehicle track mode types, and identifying normal vehicle tracks and abnormal vehicle tracks;
step 4, extracting a normal vehicle track in a single mode based on the vehicle track type analyzed in the step 3, further analyzing the vehicle track in the single mode, and identifying an abnormal vehicle track and a normal vehicle track in the single mode;
And 5, analyzing road conditions based on the abnormal vehicle tracks in the steps 3 and 4.
2. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering according to claim 1, characterized in that: in the step 1, the method specifically comprises the following steps:
step 1-1, collecting the type of road junction and channelized information thereof;
Step 1-2, extracting original track points according to different vehicle IDs based on the road monitoring video in a unit time period, and marking the original track points as P (f, x, y), wherein f represents the frame number, and x and y represent the coordinate values of the track points;
step 1-3, original track points of the vehicle are drawn in two-dimensional coordinates based on the original track points P, important characteristic data and incorrect data are determined to be missing and removed, and original track point cleaning is achieved; the missing important characteristic data is deviation data which can form a short path but deviates from the line trend, and incorrect data can not form a dense scatter set which can not be smoothly connected and can not form a path track.
3. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering according to claim 1, characterized in that: in the step 2, the method specifically comprises the following steps:
step 2-1, analyzing the original track data P based on the same vehicle ID, and extracting quantitative data from unit time within a statistical time period T to realize track feature point extraction;
Step 2-2, drawing tracks based on the extracted feature points, arranging according to the number of f frames, determining vehicle tracks, and recording as TR ═ P | PiI is more than or equal to 1 and less than or equal to n, and n is the number of track points }.
4. the intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering according to claim 1, characterized in that: in the step 3, the method specifically comprises the following steps:
Step 3-1, controlling a similarity threshold gamma, determining the range of the cluster number K, and determining the optimal cluster number K under the similarity threshold gamma based on a hierarchical clustering algorithmopiAnd its clustering result;
Step 3-2, repeating the step 3-1, adjusting the similarity threshold gamma, and repeatedly solving the optimal clustering cluster number K under different similarity thresholds gammaopiIs marked aswherein l represents a similarity threshold number, wherein the similarity threshold γlTaking an integer from the numerical value; at the same time eachMean intra-group distance with optimal cluster numberAnd group inner distance
step 3-3 based onSolving different similarity threshold values gammalMean square difference in groupAnd mean square error between groupsaccording to mean square difference in groupand mean square error between groupsDetermining an optimal threshold value gammaopiand the number of optimal groups Kopi；
Step 3-4, based on the optimal threshold value gammaopiAnd the optimal cluster number KopiDetermining a clustering result of the vehicle track, identifying a vehicle track mode type, and determining a normal vehicle track and an abnormal vehicle track;
Specifically, the optimal threshold γ is determined based on step 3-3opiAnd the optimal cluster number Kopiand (3) carrying out vehicle track hierarchical clustering, classifying the clustered vehicle track modes according to the intersection canalization information in the step (1), defaulting the vehicle tracks except for downward division according to the intersection canalization flow direction as abnormal vehicle tracks, and taking the rest as normal vehicle tracks.
5. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering of claim 4, wherein: in the step 3-1, the method comprises the following steps:
Step 3-1-1, controlling a similarity threshold gamma, and determining the longest common subsequence and the longest common subsequence similarity distance between every two tracks through an LCSS algorithm;
Step 3-1-2, listing a similarity matrix S [ a ] [ b ] according to the longest common subsequence similarity distance in the previous step, namely an adjacent matrix;
Step 3-1-3, determining the range of the cluster number K; specifically, determining the minimum value of the clustering cluster number K according to the intersection channelized information in the step 1;
step 3-1-4, giving a cluster number K, and continuously repeating the step 3-1-1 and the step 3-1-2 to obtain a hierarchical clustering result;
Step 3-1-5, determining clustering conditions with different control similarity threshold gamma and timing K values according to the cluster range given in the step 3-1-3, and establishing an evaluation system according to the intra-group distances and inter-group distances under the clustering of different K values, thereby determining the optimal clustering cluster number KopiThe method comprises the following steps:
Extracting all characteristic points P of the tracks in the group according to the clustering result of the step 3-1-4iAccording to the characteristic point PiDetermining K cluster centers { C ] under the condition of determining K groups of partitions1，C2，C3，...，CkK is more than 1 and less than or equal to K, wherein the center C of the clusterkis the center of all feature points in the group;
Respectively solving the group inner distance tau in k groups according to the k group cluster centerskAnd inter-group spacing
Group internal distance mean value based on different group numbers KAnd distance between groupsDetermining an optimal number of clusters Kopispecifically, the ratio isThe maximum K value defaults to Kopi。
6. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering of claim 4, wherein: in the step 3-3, the method comprises the following steps:
step 3-3-1, according toSolving different similarity threshold values gammalMean square difference in group
Step 3-3-2, according toSolving different similarity threshold values gammalMean square error between groups
Step 3-3-3, based on different similarity threshold values gamma and K thereofopimean square difference in groupAnd mean square error between groupsDetermining an optimal threshold value gammaopi. Specifically, the ratio isGamma at maximum timelDefaults to an optimal threshold.
7. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering according to claim 1, characterized in that: in the step 4, the method specifically comprises the following steps:
Step 4-1, extracting a normal vehicle track in a single mode, and based on the track characteristic point numerical value of the normal vehicle track, using the track similarity lambda and the acceleration variance alpha of the track2taking the arc length ratio sigma as a characteristic numerical value to perform hierarchical clustering, and dividing an abnormal driving track, an abnormal behavior track and a normal vehicle track;
Step 4-2, integrating the normal vehicle tracks obtained by analysis in the previous step, performing hierarchical clustering on the normal vehicle tracks in the previous step again under a DTW (dynamic time warping) algorithm, and judging normal vehicle tracks and outlier tracks;
And 4-3, repeating the steps 4-1 to 4-2, analyzing the vehicle track in each mode of the road junction, and dividing the abnormal driving track, the abnormal behavior track, the outlier track and the normal vehicle track in each flow direction, wherein the abnormal driving track, the abnormal behavior track and the outlier track are all defaulted as the abnormal vehicle track.
8. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering of claim 7, wherein: in the step 4-1, the method specifically comprises the following steps:
Step 4-1-1, vehicle track TRjTrack similarity lambda and acceleration variance alpha2Solving the arc length ratio sigma;
Determining a similarity calculation formula, such as an LCSS algorithm or a DTW algorithm, according to the vehicle track characteristics to obtain a vehicle track similarity lambda;
Determining acceleration variance alpha of each vehicle track based on track characteristic points of vehicle tracks TR2Namely:
In the formula, alphai+1Represents the acceleration of the characteristic point i +1, where fiand fi+1Which represents the number of frames,Denotes the Euclidean distance, p, between the feature point i and the feature point i +1iand pi+1Representing the feature points;
Determining a vehicle trajectory arc length ratio σ based on the vehicle trajectory TR, i.e.:
In the formula, p1、pi、pi+1、pnEach represents a characteristic point within the vehicle trajectory;Representing the Euclidean distance between the vehicle track characteristic point i and the characteristic point i + 1;Representing the Euclidean distance between the vehicle track characteristic point n and the characteristic point 1;
Step 4-1-2, similarity lambda and acceleration variance alpha based on the vehicle track TR in the previous step2Taking the arc length ratio sigma value as characteristic data to perform hierarchical clustering, and dividing the characteristic data into an abnormal driving track, an abnormal behavior track and a normal vehicle track; specifically, the degree of similarity λ and the acceleration variance α are used2Taking the arc length ratio sigma value as characteristic data to perform hierarchical clustering, dividing the vehicle track into three groups of data, and determining the data type according to the data quantity of the three groups of data and the discrete degree of each group of data;
Calculating the track number of the three groups of vehicle tracks and the ratio of the track number to the total track number, and if the track number in the group is the least, the data of the group of vehicle tracks is defaulted as an abnormal running track;
Respectively solving the dispersion degree epsilon of the rest two groups of vehicle track data based on the characteristic points of the vehicle tracks, namely:
In the formula, N is the total number of all vehicle track characteristic points in the group; p is a radical ofiin order to be a characteristic point, the method comprises the following steps of,Is the cluster center point;represents piDistance from point to cluster center point;
And defaulting the group of data with larger dispersion degree epsilon as an abnormal behavior track, defaulting the group of data with smaller dispersion degree epsilon as a normal vehicle track, and simultaneously realizing normal vehicle track extraction.
9. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering according to claim 1, characterized in that: in the step 5, the method specifically comprises the following steps:
step 5-1, integrating the abnormal vehicle tracks in the steps 3 and 4, analyzing the reasons for the abnormal vehicle tracks, meanwhile, corresponding the abnormal vehicle tracks to the abnormal reasons, and establishing an abnormal track association database;
And 5-2, analyzing the road junction canalization organization problem and the traffic signal control problem based on the abnormal track number proportion and the behavior reason thereof in the statistical time period, and identifying unreasonable road junction canalization and unreasonable road junction signal schemes.
10. The intersection abnormal vehicle track recognition and analysis method based on hierarchical clustering of claim 9, wherein: in the step 5-2, the concrete steps are as follows:
If the number of the abnormal tracks in the statistical time period is larger than the threshold value of the abnormal condition of the intersection, analyzing whether canalization has a problem or not;
If the same flow is downward, counting the number of abnormal tracks in the time period release phase stage, and if the number of abnormal tracks is greater than the threshold value of the abnormal condition, analyzing whether the signal scheme configuration of the phase stage is reasonable.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911012550.4A CN110570658B (en) | 2019-10-23 | 2019-10-23 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
| PCT/CN2020/095098 WO2021077761A1 (en) | 2019-10-23 | 2020-06-09 | Intersection abnormal vehicle trajectory identification and analysis method based on hierarchical clustering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911012550.4A CN110570658B (en) | 2019-10-23 | 2019-10-23 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110570658A true CN110570658A (en) | 2019-12-13 |
| CN110570658B CN110570658B (en) | 2022-02-01 |
Family
ID=68785768
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911012550.4A Active CN110570658B (en) | 2019-10-23 | 2019-10-23 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110570658B (en) |
| WO (1) | WO2021077761A1 (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110941278A (en) * | 2019-12-20 | 2020-03-31 | 交控科技股份有限公司 | In-station dynamic security analysis method |
| CN111539454A (en) * | 2020-03-30 | 2020-08-14 | 武汉理工大学 | Vehicle track clustering method and system based on meta-learning |
| CN111582317A (en) * | 2020-04-15 | 2020-08-25 | 扬州哈工科创机器人研究院有限公司 | Road positioning method and device |
| CN111652112A (en) * | 2020-05-29 | 2020-09-11 | 北京百度网讯科技有限公司 | Lane flow direction identification method and device, electronic equipment and storage medium |
| CN111681417A (en) * | 2020-05-14 | 2020-09-18 | 北京百度网讯科技有限公司 | Traffic intersection canalization adjusting method and device |
| CN111724599A (en) * | 2020-06-30 | 2020-09-29 | 暨南大学 | Method, device, equipment and medium for acquiring safe driving behavior evaluation data |
| CN111882873A (en) * | 2020-07-22 | 2020-11-03 | 平安国际智慧城市科技股份有限公司 | Track anomaly detection method, device, equipment and medium |
| CN111968365A (en) * | 2020-07-24 | 2020-11-20 | 武汉理工大学 | Non-signalized intersection vehicle behavior analysis method and system and storage medium |
| CN112307286A (en) * | 2020-11-09 | 2021-02-02 | 西南大学 | Vehicle track clustering method based on parallel ST-AGNES algorithm |
| CN112309126A (en) * | 2020-10-30 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | License plate detection method and device, electronic equipment and computer readable storage medium |
| CN112396224A (en) * | 2020-11-13 | 2021-02-23 | 智邮开源通信研究院(北京)有限公司 | Trajectory-based vehicle recruitment method, system, device and readable storage medium |
| WO2021077761A1 (en) * | 2019-10-23 | 2021-04-29 | 江苏智通交通科技有限公司 | Intersection abnormal vehicle trajectory identification and analysis method based on hierarchical clustering |
| CN113837512A (en) * | 2020-06-23 | 2021-12-24 | 中国移动通信集团辽宁有限公司 | Abnormal user identification method and device |
| CN114021629A (en) * | 2021-10-26 | 2022-02-08 | 之江实验室 | Vehicle track motion mode extraction method based on mean value dynamic time warping |
| CN115880898A (en) * | 2022-11-10 | 2023-03-31 | 公安部交通管理科学研究所 | Method for judging abnormal use of private new energy automobile |
| CN116386339A (en) * | 2023-06-06 | 2023-07-04 | 中路高科交通科技集团有限公司 | Intelligent detection method for illegal running behavior of inter-provincial passenger transport not according to line |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113469075A (en) * | 2021-07-07 | 2021-10-01 | 上海商汤智能科技有限公司 | Method, device and equipment for determining traffic flow index and storage medium |
| CN114170057A (en) * | 2021-11-17 | 2022-03-11 | 中国地质大学（武汉） | Three-dimensional intersection fine road information acquisition method and device based on track big data |
| CN114073625B (en) * | 2021-12-13 | 2023-12-08 | 曲阜师范大学 | Autonomous navigation's electronic wheelchair of brain electricity control |
| CN114999151B (en) * | 2022-05-24 | 2023-01-24 | 电子科技大学 | Density-based urban traffic flow hierarchical analysis method and device in GPS track |
| CN115631629B (en) * | 2022-10-20 | 2023-11-24 | 湖南大学 | Urban dynamic vehicle cloud construction method and system based on track prediction |
| CN116092037B (en) * | 2023-02-13 | 2023-07-28 | 长沙理工大学 | Vehicle type identification method integrating track space-semantic features |
| CN115798724B (en) * | 2023-02-13 | 2023-04-11 | 深圳市双佳医疗科技有限公司 | Human body index abnormity analysis method based on human body noninvasive data collection |
| CN116168538B (en) * | 2023-02-23 | 2024-04-12 | 公安部交通管理科学研究所 | Planar road vehicle passing space identification method |
| CN116662864B (en) * | 2023-06-14 | 2024-04-23 | 同济大学 | Rolling data cleaning method for online water quality and hydrodynamic force monitoring data |
| CN117349688B (en) * | 2023-12-01 | 2024-03-19 | 中南大学 | Track clustering method, device, equipment and medium based on peak track |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107103754A (en) * | 2017-05-10 | 2017-08-29 | 华南师范大学 | A kind of road traffic condition Forecasting Methodology and system |
| CN107240264A (en) * | 2017-06-26 | 2017-10-10 | 安徽四创电子股份有限公司 | A kind of non-effective driving trace recognition methods of vehicle and urban road facility planing method |
| CN108571975A (en) * | 2017-03-07 | 2018-09-25 | 现代自动车株式会社 | Vehicle and its control method and the autonomous driving system for using the vehicle and method |
| CN108665704A (en) * | 2018-05-17 | 2018-10-16 | 北京航空航天大学 | Intersection period maximum queue length method of estimation based on crowdsourcing track data |
| CN108986453A (en) * | 2018-06-15 | 2018-12-11 | 华南师范大学 | A kind of traffic movement prediction method based on contextual information, system and device |
| CN109242024A (en) * | 2018-09-13 | 2019-01-18 | 中南大学 | A kind of vehicle behavior similarity calculating method based on bayonet data |
| CN109410586A (en) * | 2018-12-13 | 2019-03-01 | 中南大学 | A kind of Traffic State Detection Method based on multivariate data fusion |
| US20190304298A1 (en) * | 2018-01-23 | 2019-10-03 | International Business Machines Corporation | Detection of vehicle queueing events on a road |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI430212B (en) * | 2010-06-08 | 2014-03-11 | Gorilla Technology Inc | Abnormal behavior detection system and method using automatic classification of multiple features |
| US20130286198A1 (en) * | 2012-04-25 | 2013-10-31 | Xerox Corporation | Method and system for automatically detecting anomalies at a traffic intersection |
| CN104504897B (en) * | 2014-09-28 | 2017-10-31 | 北京工业大学 | A kind of analysis of intersection traffic properties of flow and vehicle movement Forecasting Methodology based on track data |
| CN106383868A (en) * | 2016-09-05 | 2017-02-08 | 电子科技大学 | Road network-based spatio-temporal trajectory clustering method |
| CN109840660B (en) * | 2017-11-29 | 2021-07-30 | 北京四维图新科技股份有限公司 | Vehicle characteristic data processing method and vehicle risk prediction model training method |
| CN110570658B (en) * | 2019-10-23 | 2022-02-01 | 江苏智通交通科技有限公司 | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering |
-
2019
- 2019-10-23 CN CN201911012550.4A patent/CN110570658B/en active Active
-
2020
- 2020-06-09 WO PCT/CN2020/095098 patent/WO2021077761A1/en active Application Filing
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108571975A (en) * | 2017-03-07 | 2018-09-25 | 现代自动车株式会社 | Vehicle and its control method and the autonomous driving system for using the vehicle and method |
| CN107103754A (en) * | 2017-05-10 | 2017-08-29 | 华南师范大学 | A kind of road traffic condition Forecasting Methodology and system |
| CN107240264A (en) * | 2017-06-26 | 2017-10-10 | 安徽四创电子股份有限公司 | A kind of non-effective driving trace recognition methods of vehicle and urban road facility planing method |
| US20190304298A1 (en) * | 2018-01-23 | 2019-10-03 | International Business Machines Corporation | Detection of vehicle queueing events on a road |
| CN108665704A (en) * | 2018-05-17 | 2018-10-16 | 北京航空航天大学 | Intersection period maximum queue length method of estimation based on crowdsourcing track data |
| CN108986453A (en) * | 2018-06-15 | 2018-12-11 | 华南师范大学 | A kind of traffic movement prediction method based on contextual information, system and device |
| CN109242024A (en) * | 2018-09-13 | 2019-01-18 | 中南大学 | A kind of vehicle behavior similarity calculating method based on bayonet data |
| CN109410586A (en) * | 2018-12-13 | 2019-03-01 | 中南大学 | A kind of Traffic State Detection Method based on multivariate data fusion |
Non-Patent Citations (1)
| Title |
|---|
| 韩旭: "基于车辆轨迹多特征的聚类分析及异常检测方法的研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021077761A1 (en) * | 2019-10-23 | 2021-04-29 | 江苏智通交通科技有限公司 | Intersection abnormal vehicle trajectory identification and analysis method based on hierarchical clustering |
| CN110941278A (en) * | 2019-12-20 | 2020-03-31 | 交控科技股份有限公司 | In-station dynamic security analysis method |
| CN111539454A (en) * | 2020-03-30 | 2020-08-14 | 武汉理工大学 | Vehicle track clustering method and system based on meta-learning |
| CN111539454B (en) * | 2020-03-30 | 2023-04-18 | 武汉理工大学 | Vehicle track clustering method and system based on meta-learning |
| CN111582317A (en) * | 2020-04-15 | 2020-08-25 | 扬州哈工科创机器人研究院有限公司 | Road positioning method and device |
| CN111681417A (en) * | 2020-05-14 | 2020-09-18 | 北京百度网讯科技有限公司 | Traffic intersection canalization adjusting method and device |
| US11443621B2 (en) | 2020-05-14 | 2022-09-13 | Apollo Intelligent Connectivity (Beijing) Technology Co., Ltd. | Method and apparatus for adjusting channelization of traffic intersection |
| CN111652112A (en) * | 2020-05-29 | 2020-09-11 | 北京百度网讯科技有限公司 | Lane flow direction identification method and device, electronic equipment and storage medium |
| CN111652112B (en) * | 2020-05-29 | 2023-06-30 | 阿波罗智联(北京)科技有限公司 | Lane flow direction identification method and device, electronic equipment and storage medium |
| CN113837512A (en) * | 2020-06-23 | 2021-12-24 | 中国移动通信集团辽宁有限公司 | Abnormal user identification method and device |
| CN111724599A (en) * | 2020-06-30 | 2020-09-29 | 暨南大学 | Method, device, equipment and medium for acquiring safe driving behavior evaluation data |
| CN111882873A (en) * | 2020-07-22 | 2020-11-03 | 平安国际智慧城市科技股份有限公司 | Track anomaly detection method, device, equipment and medium |
| CN111968365A (en) * | 2020-07-24 | 2020-11-20 | 武汉理工大学 | Non-signalized intersection vehicle behavior analysis method and system and storage medium |
| CN112309126A (en) * | 2020-10-30 | 2021-02-02 | 杭州海康威视数字技术股份有限公司 | License plate detection method and device, electronic equipment and computer readable storage medium |
| CN112309126B (en) * | 2020-10-30 | 2022-07-05 | 杭州海康威视数字技术股份有限公司 | License plate detection method and device, electronic equipment and computer readable storage medium |
| CN112307286A (en) * | 2020-11-09 | 2021-02-02 | 西南大学 | Vehicle track clustering method based on parallel ST-AGNES algorithm |
| CN112396224A (en) * | 2020-11-13 | 2021-02-23 | 智邮开源通信研究院(北京)有限公司 | Trajectory-based vehicle recruitment method, system, device and readable storage medium |
| CN114021629A (en) * | 2021-10-26 | 2022-02-08 | 之江实验室 | Vehicle track motion mode extraction method based on mean value dynamic time warping |
| CN114021629B (en) * | 2021-10-26 | 2024-04-05 | 之江实验室 | Vehicle track motion mode extraction method based on mean dynamic time warping |
| CN115880898A (en) * | 2022-11-10 | 2023-03-31 | 公安部交通管理科学研究所 | Method for judging abnormal use of private new energy automobile |
| CN115880898B (en) * | 2022-11-10 | 2024-05-10 | 公安部交通管理科学研究所 | Private new energy automobile abnormal use judging method |
| CN116386339A (en) * | 2023-06-06 | 2023-07-04 | 中路高科交通科技集团有限公司 | Intelligent detection method for illegal running behavior of inter-provincial passenger transport not according to line |
| CN116386339B (en) * | 2023-06-06 | 2023-08-15 | 中路高科交通科技集团有限公司 | Intelligent detection method for illegal running behavior of inter-provincial passenger transport not according to line |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021077761A1 (en) | 2021-04-29 |
| CN110570658B (en) | 2022-02-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570658B (en) | Method for identifying and analyzing abnormal vehicle track at intersection based on hierarchical clustering | |
| CN110728842B (en) | Abnormal driving early warning method based on reasonable driving range of vehicles at intersection | |
| CN110717433A (en) | Deep learning-based traffic violation analysis method and device | |
| CN108596202B (en) | Method for calculating personal commuting time based on mobile terminal GPS positioning data | |
| CN111210612B (en) | Method for extracting bus route track based on bus GPS data and station information | |
| CN111243277A (en) | Commuting vehicle space-time trajectory reconstruction method and system based on license plate recognition data | |
| CN111818313B (en) | Vehicle real-time tracking method and device based on monitoring video | |
| CN106651027B (en) | Internet regular bus route optimization method based on social network | |
| CN109191872B (en) | Intersection traffic flow characteristic parameter extraction method based on number plate data | |
| CN106323301A (en) | Road information obtaining method and device | |
| CN107301776A (en) | Track road conditions processing and dissemination method based on video detection technology | |
| CN110807919A (en) | Urban road network traffic operation situation evaluation method based on vehicle passing data | |
| CN109615851B (en) | Sensing node selection method based on key road section in crowd sensing system | |
| CN102567380A (en) | Method for searching vehicle information in video image | |
| CN112365711A (en) | Vehicle track reconstruction method based on license plate recognition data | |
| CN110634299A (en) | Urban traffic state fine division and identification method based on multi-source track data | |
| CN110705484B (en) | Method for recognizing continuous lane change illegal behaviors by utilizing driving track | |
| CN107818332B (en) | Expressway interchange service range analysis method and device | |
| CN107590999A (en) | A kind of traffic state judging method based on bayonet socket data | |
| CN112991755A (en) | Method and system for selecting based on identifying commuting vehicle and constructing travel path | |
| CN112216113A (en) | Traffic flow data clustering method based on density peak value | |
| CN114372503A (en) | Cluster vehicle motion trail prediction method | |
| CN114091581A (en) | Vehicle operation behavior type identification method based on sparse track | |
| CN116168538B (en) | Planar road vehicle passing space identification method | |
| CN109633716B (en) | GPS-based urban distribution vehicle travel chain and characteristic identification method and equipment thereof |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |