WO2021077760A1

WO2021077760A1 - Abnormal driving early warning method on basis of reasonable driving range of vehicle at intersection

Info

Publication number: WO2021077760A1
Application number: PCT/CN2020/095096
Authority: WO
Inventors: 吕伟韬; 张子龙; 周东; 李璐
Original assignee: 江苏智通交通科技有限公司
Priority date: 2019-10-23
Filing date: 2020-06-09
Publication date: 2021-04-29
Also published as: CN110728842B; CN110728842A

Abstract

An abnormal driving early warning method on the basis of the reasonable driving range of a vehicle at an intersection. The method uses historical surveillance video data as a basis, comprehensively analyzes vehicle trajectories at intersections by means of a hierarchical algorithm, an LCSS algorithm and a DTW algorithm, and determines a reasonable vehicle trajectory range in each flow direction of the intersection, thus achieving prompt abnormal driving vehicle early warning on the basis of the surveillance video, effectively determining illegal behavior, promptly learning of intersection abnormalities, improving traffic management efficiency, enhancing the strength of black dot regulation and control at intersections, and providing a safe and clear traffic environment for intersections.

Description

Abnormal driving early warning method based on reasonable driving range of vehicles at intersection

Technical field

The invention relates to the field of vehicle driving monitoring, in particular to an abnormal driving early warning method based on a reasonable driving range of a vehicle at an intersection.

Background technique

With the acceleration of urbanization and the continuous increase of car ownership, urban road traffic conditions have become very complicated. In the urban road network, intersections, as the "throat" of urban traffic, are the core of the entire urban traffic. Whether vehicles at intersections drive in order and in accordance with regulations is a key factor leading to traffic accidents and traffic congestion. Therefore, how to drive vehicles at intersections Trajectory identification and realizing early warning of illegal vehicles are one of the key points to alleviate urban traffic problems today.

At the current stage of the development of intelligent transportation, the popularization of electronic police intelligent equipment provides effective support conditions for illegal driving of vehicles at intersections. For example, the invention CN201520771567.9 proposed a capture system for illegal driving of vehicles. Hardware manufacturers such as Hisense and Dahua have adopted The detection equipment of the road network and the electronic police realize off-site law enforcement against violations such as reverse driving.

At present, the illegal driving of vehicles is only an early warning and off-site law enforcement for "line pressing", "reverse" and other violations. However, in actual vehicle operation, "left-turning vehicles use the straight lane to bypass" and other "drilling holes" often occur. "Illegal behaviors or abnormal detours at intersections, these abnormal driving will aggravate intersection congestion and accidents.

Summary of the invention

In view of the above-mentioned technical problems, the present invention proposes an abnormal driving early warning method based on the reasonable driving range of intersection vehicles. The method takes the intersection as the research object and realizes the extraction and analysis of the intersection vehicle trajectory based on historical monitoring video data, and effectively distinguishes the abnormal vehicle trajectory Realize the extraction and analysis of normal vehicle trajectories, and then determine the reasonable driving range of vehicles in each flow direction of the intersection according to the normal vehicle trajectory, establish a normal range database, and provide early warning for real-time violation vehicles and abnormal behavior at intersections, thereby improving the efficiency of intersection supervision and preventing traffic accidents at intersections. Congestion provides effective preventive measures.

The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle includes the following steps:

Step 1: Determine the trajectory of each vehicle at the intersection based on the historical video of the intersection monitoring, and classify the vehicle trajectory type pattern through hierarchical clustering and intersection channelization;

Step 2: Extract and analyze the reasonable range of vehicle trajectories at the intersections, and realize the management of the reasonable range of vehicles at the intersection;

Step 3: Early warning of past abnormal driving vehicles based on the real-time monitoring video information of the intersection to realize the monitoring and management of the abnormal situation at the intersection.

Further, in the step 1, the specific steps are as follows:

Step 1-1, based on the historical video of the intersection monitoring unit time period, extract the original trajectory points under different vehicle IDs, denoted as P(f,x,y), where f represents the number of frames, x and y represent the coordinate values, and then pass determining the initial data of different cleaning vehicle ID of a single track vehicle, referred to as _{TR = {P | p i,} 1≤i≤n, n } is the locus of points;

Step 1-2, classify the vehicle trajectory type pattern based on the hierarchical clustering algorithm and the intersection channelization information. Specifically, the feature points p _{i are} extracted from a single vehicle trajectory, and the vehicle trajectory hierarchical clustering is performed through Euclidean distance or LCSS distance, and abnormal vehicle trajectories are eliminated, and vehicle trajectories are divided according to the characteristics of intersection channelization.

Further, the step 2 specifically includes the following steps:

Step 2-1: Extract the vehicle trajectory TR of a single mode type, and realize the extraction and analysis of the normal vehicle trajectory in this mode type through hierarchical clustering.

Step 2-2: Perform trajectory smoothing based on the normal vehicle trajectory extracted by screening in step 2-1, re-extract feature points of the vehicle trajectory, and determine the single-directional vehicle operating range.

Step 2-3, repeat steps 2-1 and 2-2 to determine the envelope trajectory of each flow direction at the intersection, and integrate the reasonable driving range of the vehicle trajectory at the intersection.

Further, the step 2-1 specifically includes the following sub-steps:

In step 2-1-1, the trajectory similarity λ, variance α ² and arc length ratio σ of a single vehicle trajectory based on the feature points p _i _{of the vehicle trajectory TR j are solved as feature values.} Among them, 1≤i≤n, n is the number of trajectory feature points; 1≤j≤N, N is the number of vehicle trajectories; specifically, according to the type of vehicle trajectory, the similarity _{of vehicle trajectory TR j can be obtained through LCSS algorithm or DTW algorithm} [lambda]; the distance between the frames and the feature points of the vehicle trajectory TR _j p _i is determined based on the variance of the acceleration of the vehicle track TR _j α ^2; determining the vehicle trajectory TR _j arc distance between the vehicle based on the feature point trajectory TR _j Length ratio σ;

Step 2-1-2, use the similarity λ, acceleration variance α ² and arc length ratio σ _{j of the} _{vehicle trajectory TR j} as feature data to perform hierarchical clustering of the vehicle trajectory, and divide it into abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory Three types, and then eliminate the abnormal driving trajectory and abnormal behavior trajectory, realize the normal vehicle trajectory data extraction; specifically, integrate the similarity λ _j and acceleration variance of _{each vehicle trajectory TR j}

The arc length ratio σ _j value is used as the characteristic value, and the hierarchical clustering algorithm is used to divide the data into three groups. According to the data volume and the degree of dispersion ε of each group of data, the abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory are determined. Among them, the group of data with the smallest amount of data and the ratio of the number of the group to the total number is less than the abnormal threshold is defaulted as the abnormal trajectory, and the group of data with a larger degree of dispersion ε is defaulted as the abnormal behavior trajectory;

Step 2-1-3, based on the DTW algorithm, re-cluster the normal vehicle trajectory of step 2-1-2, and distinguish between the normal vehicle trajectory and the outlier trajectory based on the number of trajectories and the degree of dispersion ε of each cluster after clustering , To extract the normal vehicle trajectory.

Further, the step 2-2 specifically includes the following sub-steps:

Step 2-2-1, filter and extract the trajectory points p _i _{under the normal vehicle trajectory TR j} to achieve smooth vehicle trajectory, and select new feature points p′ _{i to} determine the vehicle trajectory according to the smoothed vehicle trajectory points equally;

_{Step 2-2-2:} Determine the main curve of vehicle operation based on all vehicle trajectories TR _j and their new feature points p′ i; specifically, based on the selected feature points p′ _{i of} each vehicle trajectory, correspond them one by one , The center point of stage i is determined under the same stage i, and the main curve TR ^mid is obtained, namely:

Wherein, TR ^mid master curve showing the trajectory of the vehicle; P _i ^mid central point of the i-th stage indicates trajectories of the feature points, wherein 1≤i≤n, n represents the number of feature points selected; p _'ji represents the j-th track vehicle The new feature points of the i-th stage trajectory, where 1≤j≤N, N represents the total number of vehicle trajectories;

At the same time, based on the numerical distribution of the distance between each stage and the center point, determine the outliers of each stage, namely:

In the formula, dist(p' _ji , P _i ^mid ) represents the distance between the characteristic point of j trajectories and the center point of stage i;

Analyze the distance values of all the trajectory points in the i stage, and through the normal distribution analysis, the trajectory stage points corresponding to the values greater than the normal distribution (μ+2σ) are defaulted as outliers;

Step 2-2-3, in accordance with the master curve and master curves TR ^mid point P _i ^mid stage of each stage is determined envelope points Q _i, and thus obtained two smooth envelope, determines a vehicle trajectory with a reasonable range;

Further, the steps 2-2-3 specifically include the following sub-steps:

Step 2-2-3-1: Determine the distance between the vehicle trajectory point p′ _ji and the main curve TR ^mid in the same phase according to the stage point P _i ^{mid on the} main curve, where the closest point to the main curve is not on the main curve TR. ^{At mid} , it means that the vehicle trajectory point has shifted at this stage, and it is removed, otherwise, go to the next step;

Steps 2-2-3-2, the trajectory of the vehicle deviates from the respective track point does not occur at the same stage p _'ji main curve TR ^mid distance _{^{dist (p' ji, TR mid}} ) sorting, selecting 80% -90% The median value is taken as the radius distance of the main curve corresponding to the envelope curve at this stage, denoted as r _i ;

2-2-3-3 step, based on the main stage of the point P _i ^mid curve determined in the same phase envelope coordinates Q _i, denoted _{_{(x 'i, y' i}} ), specifically,

y′ _i =k _i x′ _i +b _i

Wherein, x _'i and y' _i represents the coordinates of the package on the envelope stage i, denoted by Q _i,

with

Curvilinear coordinates on a front stage i, denoted as P _{_i} ^mid; k _i represents the i-th stage of inclination of the envelope bag, B _i represents the lower envelope curve of the main lines of intersection intercept i stage;

Step 2-2-3-4, judge the position relationship between it and the main curve according to the envelope coordinates solved in the previous step; specifically, according to the main curve point P _i ^{mid of} this stage and the main curve point of the next stage

Obtain the envelope threshold ξ, namely:

In the formula, x′ _i and y′ _i represent the point coordinates of the envelope on the i stage;

with

Indicates the coordinates of the master curve on stage i;

If ξ<0, the _{envelope points represented by x′ i} and y′ _i are on the left side of the main curve, and ξ>0 indicates that the _{envelope points of x′ i} and y′ _i are on the right side of the main curve, and vice versa. On the straight line where the line intersects the main curve;

Step 2-2-3-5: Integrate the envelope point coordinates Q _{i of} all the stage points, smooth the envelope trajectory, and obtain a new envelope characteristic point q _s (i), thereby determining the envelope trajectory TR ^q , the range composed of two lines is the reasonable driving range of the vehicle trajectory in this mode.

Further, the step 3 includes the following sub-steps:

Step 3-1: Determine the direction and track information of the vehicle based on real-time surveillance video analysis;

Step 3-2, compare the vehicle trajectory path obtained in step 3-1 with the reasonable driving range of the intersection vehicle trajectory obtained in step 2; if the vehicle exceeds the driving range, an early warning is given and the vehicle information is sent to the off-site law enforcement system ；

Step 3-3: If the ratio between the number of vehicles running downward and the total number of vehicles passing in a unit time is greater than the threshold of abnormal conditions, the intersection flow direction will be warned, and it can be linked to the traffic situation monitoring system or video monitoring system at the same time. Alarm for abnormal conditions at intersections to monitor abnormal conditions at intersections.

The beneficial effect achieved by the present invention is as follows: Compared with the traditional GPS vehicle trajectory clustering abnormal analysis mode, the present invention is based on historical video data, and realizes the extraction and analysis of the vehicle trajectory at the intersection through the hierarchical clustering algorithm, and effectively combines the abnormal vehicle trajectory and abnormal behavior trajectory. , Outlier trajectory and other vehicle trajectory recognition, and extract the correct vehicle trajectory. Compared with the traditional early warning method of illegal and abnormal intersections, the present invention extracts and analyzes the correct vehicle trajectory, determines the reasonable driving range of vehicles in each direction of the intersection, and compares the original intersection trajectory interval to a smaller and more reasonable interval, and effectively warns abnormal vehicles driving and improves the non-compliance. The efficiency of on-site law enforcement reduces the workload of manual review and provides reference indicators for intelligent identification. At the same time, it can distinguish abnormal conditions of intersections based on abnormalities of multiple vehicles, realize remote linkage monitoring, and improve the efficiency of intersection traffic supervision.

Description of the drawings

Fig. 1 is a schematic diagram of the steps of the abnormal driving early warning method of the present invention.

Fig. 2 is a schematic diagram of a vehicle trajectory pattern divided by hierarchical clustering in an embodiment of the invention.

Fig. 3 is a schematic diagram of a trajectory after extraction of a vehicle trajectory at a single intersection in an embodiment of the present invention.

Detailed ways

The technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings of the specification.

Step 1: Determine the trajectory of each vehicle at the intersection based on the historical video of the intersection monitoring, and classify the vehicle trajectory type pattern through hierarchical clustering and intersection channelization.

Step 1-1, based on the historical video of the intersection monitoring unit time period, extract the original trajectory points under different vehicle IDs, denoted as P(f,x,y), where f represents the number of frames, x and y represent the coordinate values, and then pass determining the initial data cleaning different single track vehicle ID of the vehicle, referred to as _{TR = {P | p i,} 1≤i≤n, n is the locus of points}; in general, taking the unit time period 5min.

Step 1-2, classify the vehicle trajectory type pattern based on the hierarchical clustering algorithm and the intersection channelization information. Specifically, the feature points p _{i are} extracted from a single vehicle trajectory, and the vehicle trajectory hierarchical clustering is performed by Euclidean distance or LCSS distance, and abnormal vehicle trajectories (such as staying at intersections, driving in reverse, etc.) Channelization features realize vehicle trajectory division. Generally speaking, taking an intersection as an example, the types of vehicle trajectories include left, straight, and right turns in four directions: east, south, west, and north, with a total of 12 types.

Step 2: Extract and analyze the reasonable range of vehicle trajectories at the intersections, and realize the management of the reasonable range of vehicles at the intersection.

Step 2-1, extract a single mode type vehicle trajectory TR, analyze all trajectories under this type through a hierarchical clustering algorithm, and realize normal vehicle trajectory extraction.

In step 2-1-1, the trajectory similarity λ, variance α ² and arc length ratio σ of a single vehicle trajectory based on the feature points p _i _{of the vehicle trajectory TR j are solved as feature values.} Among them, 1≤i≤n, n is the number of trajectory feature points; 1≤j≤N, N is the number of vehicle trajectories.

_{Specifically, the vehicle trajectory TR j} similarity λ is solved by the LCSS algorithm or the DTW algorithm according to the vehicle trajectory type; the vehicle trajectory TR _j is determined based on the frame number of the feature points p _i _{of the vehicle trajectory TR j} and the distance between the feature points acceleration variance α ^2; determining the vehicle trajectory TR _j σ arc length than the distance between the vehicle based on the feature point trajectory TR _j.

Among them, taking the LCSS algorithm as an example, the formula for the similarity λ of the _{vehicle trajectory TR j is:}

In the formula, TR ₁ and TR ₂ are two trajectories of length m and n respectively;

Where p _i and q _j respectively represent the coordinates of the characteristic points of the trajectory,

The same

Where 1≤i≤m, 1≤j≤n,

Indicates that the trajectory is empty; dist(p _i ,q _j ) represents the Euclidean distance between two coordinate points, and γ is the similarity threshold; LCSS(TR ₁ ,TR ₂ ) represents the longest common of the two trajectories of _{TR 1} and TR ₂ _Subsequence length; D LCSS (TR ₁ ,TR ₂ ) is the similarity distance between _{trajectories TR 1} and TR ₂ _{; min(len TR1} ,len _TR2 ) represents the smaller value of the length of _{trajectories TR 1} and TR ₂ _{; D LCSS} ( TR ₁ , TR ₁ ) represents _{the longest common subsequence similarity distance between TR 1} and TR ₁ , and N represents the number of vehicle trajectories; λ ₁ represents the similarity value of _{TR 1 vehicle trajectories.}

The acceleration variance α ² of the vehicle trajectory TR _{j is:}

In the formula, α _i+1 represents the acceleration of the feature point i+1, where f _i and f _i+1 represent the number of frames,

It represents the Euclidean distance between feature point i and feature point i+1, and p _i and p _i+1 represent feature points.

The trajectory arc length ratio σ of the vehicle trajectory TR _{j is:}

In the formula, p ₁ , p _i , p _i+1 , and p _n all represent the feature points in the vehicle trajectory TR _j;

Represents the Euclidean distance between the characteristic point i of the vehicle trajectory and the characteristic point i+1;

It represents the Euclidean distance between the feature point n of the vehicle root trajectory and the feature point 1; n represents the number of feature points in the vehicle trajectory, 1≤i≤n.

Step 2-1-2, use the similarity λ, acceleration variance α ² and arc length ratio σ _{j of the} _{vehicle trajectory TR j} as feature data to perform hierarchical clustering of the vehicle trajectory, and divide it into abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory Three types, and then eliminate the abnormal driving trajectory and abnormal behavior trajectory, realize the normal vehicle trajectory data extraction.

Specifically, integrate the similarity λ _j and acceleration variance of _{each vehicle trajectory TR j}

The arc length ratio σ _j value is used as the characteristic value, and the hierarchical clustering algorithm is used to divide the data into three groups. According to the data volume and the degree of dispersion ε of each group of data, the abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory are determined. Among them, the group of data with the smallest amount of data and the ratio of the number of the group to the total number is less than the abnormal threshold (usually 30%) is defaulted as an abnormal trajectory, and the group of data with a larger degree of dispersion ε is defaulted as an abnormal behavior, where the degree of dispersion ε is as follows :

In the formula, n'is the total number of feature points of all vehicle trajectories in the divided group; p _i is the feature point,

Is the cluster center point, that is, the group center of the divided group of data;

Indicates the _{distance from the p i} point to the cluster center point.

Step 2-1-3, under the DTW algorithm, perform hierarchical clustering again on the normal vehicle trajectory of the previous step, and distinguish the normal vehicle trajectory and the outlier trajectory.

Specifically, extract the normal vehicle trajectory in the previous step, determine the similarity between the two trajectories based on the DTW algorithm, list the similarity matrix S[a][b], and divide the vehicle trajectory into Two types include normal trajectory and outlier trajectory. The outlier trajectory is determined based on the number of trajectories or the degree of dispersion ε, so as to extract the normal vehicle trajectory.

In general, if two sets of vehicle trajectories A and B, the number of one set of trajectories is less than the number of the other set and the ratio of the number to the total number is less than the abnormal threshold (usually 30%), then the set of data is defaulted as abnormal outlier The trajectory, the other group is the normal trajectory; otherwise, the discrete degree ε of the two sets of data is solved separately, and the group of data with the larger value is defaulted as the abnormal outlier trajectory.

Step 2-2: Perform trajectory smoothing based on the vehicle trajectory extracted in step 2-1, re-extract feature points of the vehicle trajectory, and determine the vehicle operating range in a single direction.

Step 2-2-1: Screen the trajectory points p _i _{under the normal vehicle trajectory TR j} extracted in the previous step to achieve smooth vehicle trajectory, and analyze and determine the new feature point p′ _i after the vehicle trajectory is smoothed.

Specifically, methods such as moving average, weighted moving average, and exponential moving average can be used. The present invention selects the moving average smoothing method, which is specifically as follows:

In the formula, p _s (i) represents the smoothed vehicle trajectory point: ω is the smoothing index, which is determined according to the number of trajectory points of the shortest trajectory in all trajectories and the number of feature points that need to be selected. The present invention reselects 10 features Point, so the value of ω is 5.

Further, the vehicle trajectory is equally divided to re-select new feature points, which are recorded as vehicle trajectory TR={P′|p′ _i , 1≤i≤n, n is the number of trajectory feature points}. In general, ten new features are selected point.

_{Step 2-2-2:} Determine the main curve of vehicle operation based on all vehicle trajectories TR _j and their characteristic points p′ i. Specifically, based on the selected feature points p′ _{i of} each vehicle trajectory, they are one-to-one correspondence, and the center point of stage i is determined at the same stage i to obtain the main curve TR ^mid , namely:

Wherein, TR ^mid master curve showing the trajectory of the vehicle; P _i ^mid central point of the i-th stage indicates trajectories of the feature points, wherein 1≤i≤n, n represents the number of the selected feature point (in general, n-value of 10 ); p′ _ji represents the new feature points of the j-th vehicle trajectory at the i-th stage, where 1≤j≤N, and N represents the total number of vehicle trajectories.

_{Wherein, dist (p 'ji, P} i mid) represents the i th track j the distance between the center point of the feature point and stage i.

Further analyze the distance values of all the trajectory points in the i stage, and through the normal distribution analysis, the values corresponding to the trajectory stage points that are greater than the normal distribution (μ+2σ) are defaulted as outliers.

Step 2-2-3, in accordance with the master curve and master curve TR ^mid stage of a step of determining the point P _i ^mid point of each stage envelope Q _i, and thus obtain a smooth envelope.

Step 2-2-3-1: Determine the distance between the vehicle trajectory point p′ _ji and the main curve TR ^mid in the same phase according to the stage point P _i ^{mid on the} main curve, where the closest point to the main curve is not on the main curve TR. ^{At mid} , it means that the vehicle trajectory point has shifted at this stage, and it will be eliminated, otherwise, go to the next step.

Steps 2-2-3-2, the trajectory of the vehicle deviates from the respective track point does not occur at the same stage p _'ji main curve TR ^mid distance _{^{dist (p' ji, TR mid}} ) sorting, selecting 80% -90% The median value is taken as the radius distance of the envelope curve corresponding to the main curve at this stage, denoted as r _i .

2-2-3-3 step, based on the main stage of the point P _i ^mid curve determined in the same phase envelope coordinates Q _i, denoted _{_{(x 'i, y' i}} ). Specifically:

y′ _i =k _i x′ _i +b _i

with

Curvilinear coordinates on a front stage i, denoted as P _{_i} ^mid; k _i represents the slope of the i-th stage of the envelope, i B _i represents the main phase of the envelope curve intersects the line intercept.

Step 2-2-3-4, judge the positional relationship between it and the main curve according to the envelope coordinates solved in the previous step. Specifically, according to the main curve point P _i ^{mid of} this stage and the main curve point of the next stage

Obtain the envelope threshold ξ, namely:

with

Represents the coordinates of the master curve on stage i.

If ξ<0, the _{envelope points represented by x′ i} and y′ _i are on the left side of the main curve, and ξ>0 means that the _{envelope points of x i} and y _i are on the right side of the main curve. On the straight line where the main curve intersects.

Step 2-2-3-5: Integrate the envelope point coordinates Q _{i of} all the stage points, smooth the envelope trajectory, and obtain a new envelope characteristic point q _s (i), thereby determining the envelope trajectory TR ^{q is} the reasonable driving range of the vehicle trajectory in this mode (direction).

Step 2-3, repeat steps 2-1 and 2-2 to determine the envelope trajectory of each flow direction at the intersection (the reasonable driving range of the vehicle), and integrate the reasonable driving range of the vehicle trajectory at the intersection.

Specifically, taking a conventional intersection as an example, the reasonable driving range of the vehicle for the trajectory of the vehicle turning left, going straight, and turning right in the four directions is determined.

Step 3-1: Determine the direction and track information of the vehicle based on real-time surveillance video analysis.

Step 3-2, compare the vehicle trajectory path obtained in step 3-1 with the reasonable driving range of the intersection vehicle trajectory obtained in step 2; if the vehicle exceeds the driving range, an early warning is given and the vehicle information is sent to the off-site law enforcement system .

Step 3-2: If the ratio between the number of vehicles running downward and the total number of vehicles passing in a unit time is greater than the abnormal condition threshold (usually about 50%, this value can be selected according to the actual situation of the intersection), then proceed to the flow direction of the intersection Early warning can be linked to the traffic situation monitoring system or video monitoring system to realize the alarm of abnormal conditions at the intersection and monitor the abnormal conditions at the intersection.

The following uses specific examples to illustrate:

Select a certain intersection for research, extract the surveillance video within 5 minutes of the intersection, determine the vehicle trajectory according to the trajectory point, and use hierarchical clustering to divide the vehicle trajectory mode, as shown in Figure 2.

Further analyze the extraction and analysis of vehicle trajectory at a single intersection, taking a left turn at the south entrance as an example, extracting the abnormal trajectory of the vehicle, and determining the main curve and envelope according to the normal vehicle trajectory and its characteristic points, as shown in Figure 3. That is, the section between envelope 1 and envelope 2 in the figure is the reasonable left-turn driving range at the south entrance.

When a left-turning vehicle at the south entrance exceeds a reasonable driving range during passing, an early warning will be issued. If the number of early warnings in a unit time period (1 hour) exceeds 50% of the passing record, the association will be automatically prompted to realize remote monitoring of the intersection.

The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. However, all equivalent modifications or changes made by those of ordinary skill in the art based on the disclosure of the present invention should be included Within the scope of protection described in the claims.

Claims

An abnormal driving early warning method based on a reasonable driving range of vehicles at an intersection is characterized in that it includes the following steps:

Step 1: Determine the trajectory of each vehicle at the intersection based on the historical video of the intersection monitoring, and classify the vehicle trajectory type pattern through hierarchical clustering and intersection channelization;

Step 2: Extract and analyze the reasonable range of vehicle trajectories at the intersections, and realize the management of the reasonable range of vehicles at the intersection;

Step 3: Early warning of past abnormal driving vehicles based on the real-time monitoring video information of the intersection to realize the monitoring and management of the abnormal situation at the intersection.
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 1, characterized in that: in the step 1, the specific steps are as follows:

Step 1-1, based on the historical video of the intersection monitoring unit time period, extract the original trajectory points under different vehicle IDs, denoted as P(f, x, y), where f represents the number of frames, x and y represent the coordinate values, and then pass determining the initial data of different cleaning vehicle ID of a single track vehicle, referred to as TR = {P | p i, 1≤i≤n, n } is the locus of points;

Step 1-2, classify the vehicle trajectory type pattern based on the hierarchical clustering algorithm and the intersection channelization information. Specifically, the feature points p i are extracted from a single vehicle trajectory, and the vehicle trajectory hierarchical clustering is performed through Euclidean distance or LCSS distance, and abnormal vehicle trajectories are eliminated, and vehicle trajectories are divided according to the characteristics of intersection channelization.
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 1, wherein the step 2 specifically includes the following steps:

Step 2-1: Extract the vehicle trajectory TR of a single mode type, and realize the extraction and analysis of the normal vehicle trajectory in this mode type through hierarchical clustering.

Step 2-2: Perform trajectory smoothing based on the normal vehicle trajectory extracted by screening in step 2-1, re-extract feature points of the vehicle trajectory, and determine the single-directional vehicle operating range.

Step 2-3, repeat steps 2-1 and 2-2 to determine the envelope trajectory of each flow direction at the intersection, and integrate the reasonable driving range of the vehicle trajectory at the intersection.
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 3, characterized in that: the step 2-1 specifically includes the following sub-steps:

In step 2-1-1, the trajectory similarity λ, variance α 2 and arc length ratio σ of a single vehicle trajectory based on the feature points p i of the vehicle trajectory TR j are solved as feature values. Among them, 1≤i≤n, n is the number of trajectory feature points; 1≤j≤N, N is the number of vehicle trajectories; specifically, according to the type of vehicle trajectory, the similarity of vehicle trajectory TR j can be obtained through LCSS algorithm or DTW algorithm [lambda]; the distance between the frames and the feature points of the vehicle trajectory TR j p i is determined based on the variance of the acceleration of the vehicle track TR j α 2; determining the vehicle trajectory TR j arc distance between the vehicle based on the feature point trajectory TR j Length ratio σ;

Step 2-1-2, use the similarity λ, acceleration variance α 2 and arc length ratio σ j of the vehicle trajectory TR j as feature data to perform hierarchical clustering of the vehicle trajectory, and divide it into abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory Three types, and then eliminate the abnormal driving trajectory and abnormal behavior trajectory, realize the normal vehicle trajectory data extraction; specifically, integrate the similarity λ j and acceleration variance of each vehicle trajectory TR j
The arc length ratio σ j value is used as the characteristic value, and the hierarchical clustering algorithm is used to divide the data into three groups. According to the data volume and the degree of dispersion ε of each group of data, the abnormal driving trajectory, abnormal behavior trajectory and normal vehicle trajectory are determined. Among them, the group of data with the smallest amount of data and the ratio of the number of the group to the total number is less than the abnormal threshold is defaulted as the abnormal trajectory, and the group of data with a larger degree of dispersion ε is defaulted as the abnormal behavior trajectory;

Step 2-1-3, based on the DTW algorithm, re-cluster the normal vehicle trajectory of step 2-1-2, and distinguish between the normal vehicle trajectory and the outlier trajectory based on the number of trajectories and the degree of dispersion ε of each cluster after clustering , To extract the normal vehicle trajectory.
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 3, wherein the step 2-2 specifically includes the following sub-steps:

Step 2-2-1, filter and extract the trajectory points p i under the normal vehicle trajectory TR j to achieve smooth vehicle trajectory, and select new feature points p′ i to determine the vehicle trajectory according to the smoothed vehicle trajectory points equally;

Step 2-2-2: Determine the main curve of vehicle operation based on all vehicle trajectories TR j and their new feature points p′ i; specifically, based on the selected feature points p′ i of each vehicle trajectory, correspond them one by one , The center point of stage i is determined under the same stage i, and the main curve TR mid is obtained, namely:

Wherein, TR mid master curve showing the trajectory of the vehicle; P i mid central point of the i-th stage indicates trajectories of the feature points, wherein 1≤i≤n, n represents the number of feature points selected; p 'ji represents the j-th track vehicle The new feature points of the i-th stage trajectory, where 1≤j≤N, N represents the total number of vehicle trajectories;

At the same time, based on the numerical distribution of the distance between each stage and the center point, determine the outliers of each stage, namely:

Wherein, dist (p 'ji, P i mid) represents the distance between the tracks j and i stage i the center point of the feature point;

Analyze the distance values of all the trajectory points in the i stage, and through the normal distribution analysis, the trajectory stage points corresponding to the values greater than the normal distribution (μ+2σ) are defaulted as outliers;

Step 2-2-3, in accordance with the master curve and master curves TR mid point P i mid stage of each stage is determined envelope points Q i, and thus obtained two smooth envelope, determines a vehicle trajectory with a reasonable range;
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 5, wherein the step 2-2-3 specifically includes the following sub-steps:

Step 2-2-3-1: Determine the distance between the vehicle trajectory point p′ ji and the main curve TR mid in the same phase according to the stage point P i mid on the main curve, where the closest point to the main curve is not on the main curve TR. At mid , it means that the vehicle trajectory point has shifted at this stage, and it is removed, otherwise, go to the next step;

Steps 2-2-3-2, the trajectory of the vehicle deviates from the respective track point does not occur at the same stage p 'ji main curve TR mid distance dist (p' ji, TR mid ) sorting, selecting 80% -90% The median value is taken as the radius distance of the main curve corresponding to the envelope curve at this stage, denoted as r i ;

2-2-3-3 step, based on the main stage of the point P i mid curve determined in the same phase envelope coordinates Q i, denoted (x 'i, y' i ), specifically,

y′ i =k i x′ i +b i

Wherein, x 'i and y' i represents the coordinates of the package on the envelope stage i, denoted by Q i,
with
Curvilinear coordinates on a front stage i, denoted as P i mid; k i represents the i-th stage of inclination of the envelope bag, B i represents the lower envelope curve of the main lines of intersection intercept i stage;

Step 2-2-3-4, judge the position relationship between it and the main curve according to the envelope coordinates solved in the previous step; specifically, according to the main curve point P i mid of this stage and the main curve point of the next stage
Obtain the envelope threshold ξ, namely:

In the formula, x′ i and y′ i represent the point coordinates of the envelope on the i stage;
with
Indicates the coordinates of the master curve on stage i;

If ξ<0, the envelope points represented by x′ i and y′ i are on the left side of the main curve, and ξ>0 means that the envelope points of x′ i and y′ i are on the right side of the main curve, and vice versa. On the straight line where the line intersects the main curve;

Step 2-2-3-5: Integrate the envelope point coordinates Q i of all the stage points, smooth the envelope trajectory, and obtain the new envelope characteristic point q s (i), thereby determining the envelope trajectory TR q , the range composed of two lines is the reasonable driving range of the vehicle trajectory in this mode.
The abnormal driving early warning method based on the reasonable driving range of the intersection vehicle according to claim 1, wherein said step 3 includes the following sub-steps:

Step 3-1: Determine the direction and track information of the vehicle based on real-time surveillance video analysis;

Step 3-2, compare the vehicle trajectory path obtained in step 3-1 with the reasonable driving range of the intersection vehicle trajectory obtained in step 2; if the vehicle exceeds the driving range, an early warning is given and the vehicle information is sent to the off-site law enforcement system ；

Step 3-3: If the ratio between the number of vehicles running downward and the total number of vehicles passing in a unit time is greater than the threshold of abnormal conditions, the traffic direction of the intersection will be warned, and it can be linked to the traffic situation monitoring system or video monitoring system at the same time. Alarm for abnormal conditions at intersections to monitor abnormal conditions at intersections.