An intersection extraction method based on low-speed features and clustering algorithm

Abstract

The application provides an intersection extraction method based on low-speed features and a clustering algorithm, comprising the following steps: reading in trajectory data; extracting a turning pair in the trajectory, defining two continuous GPS positioning points as the turning pair, and extracting a corresponding turning point to obtain a turning point set; screening out low-speed points of all vehicle trajectories in a flat peak period; according to the number of low-speed points around each turning point, eliminating or retaining the turning point to obtain a screened turning point data set; and using a clustering algorithm based on local connectivity to classify the screened turning point data set, extract a class cluster center, and obtain an intersection; the method considers the low-speed features of a vehicle when passing through the intersection, extracts a turning point according to a heading change of the vehicle, eliminates the turning point according to the number of low-speed points, and obtains intersection information after clustering, so that the accuracy is high.

Description

TECHNICAL FIELD

[0001] The application relates to the field of traffic big data application, in particular to an intersection extraction method based on low-speed features and a clustering algorithm. BACKGROUND

[0002] With the popularity of GPS (Global Position System) devices and large-scale application in daily life, trajectory data collected by positioning devices are everywhere. The trajectory data generated by GPS positioning devices have the advantages of low cost, rich information and convenient collection, and have been widely applied to various fields in our daily life. With the rapid development of urban construction, road network information is also changing continuously. Traditional surveying and mapping and other road network information extraction technologies require a large amount of manpower and material resources, and the extraction period of road network information is too long. Vehicle GPS trajectory data have wide coverage, low cost and short period, and the extraction of road network information based on vehicle GPS trajectory data has practical significance.

[0003] An intersection is an important part of a road network structure, and the spatial position, range and detailed internal structure of a city intersection are the key to the information of a city traffic network at all levels. At present, most intersection extraction algorithms mainly include three categories: intersection extraction based on GPS trajectory data, intersection extraction based on digital images and intersection extraction based on a combination of multiple sensors. Among them, the intersection extraction methods based on GPS trajectory data mainly have the following existing schemes: intersection recognition based on local connectivity clustering algorithm, which screens turning pairs and turning points by using the different turning sizes, different turning numbers and different turning types of vehicles at intersections and non-intersections, and calculates the intersections by using the local connectivity clustering algorithm; intersection extraction based on DBSCAN clustering algorithm of multi-density zoning, which screens turning pairs and turning points by using the different turning sizes, different turning numbers and different turning types of vehicles at intersections and non-intersections, and extracts intersections by using the DBSCAN clustering algorithm of multi-density zoning; the effects of these algorithms are generally common, and some features are not considered. SUMMARY

[0004] The main purpose of the application is to overcome the above-mentioned defects in the prior art, and to provide an intersection extraction method based on low-speed features and a clustering algorithm. The method considers the low-speed features of vehicles passing through intersections during flat peak periods, extracts turning points according to the changes in the heading of vehicles, removes the turning points according to the number of low-speed points, and obtains intersection information after clustering, so that the accuracy is high.

[0005] The application adopts the following technical scheme:

[0006] An intersection extraction method based on low-speed features and a clustering algorithm comprises the following steps:

[0007] Step 1: Read in trajectory data;

[0008] Step 2: Extract the turning pairs from the trajectory. Define two consecutive GPS positioning points as a turning pair and extract the corresponding turning points to obtain the set of turning points.

[0009] Step 3: Filter out the low-speed points of all vehicle trajectories during off-peak hours;

[0010] Step 4: Based on the number of low-speed points around each turning point, remove or retain the turning point to obtain the filtered turning point dataset;

[0011] Step 5: Use a clustering algorithm based on local connectivity to classify the filtered turning point dataset and extract cluster centers to obtain the intersections.

[0012] Specifically, the method for extracting turning points is as follows:

[0013] Iterate through all the trajectories, and when the difference in heading angles between two trajectory points p1(x1,y1,k1) and p2(x2,y2,k2) is less than the turning angle threshold α... turn Furthermore, the time difference between the two trajectory points is less than the time threshold t. turn Then the two trajectory points are determined to be a turning pair, where x, y, and k are the slopes of the Cartesian coordinate system corresponding to their coordinates and headings, respectively; and using

[0014]

[0015] Calculate the turning point TP(x,y) corresponding to the steering pair, and finally obtain the set of turning points TP = {TP1, TP2, ..., TP}. n}

[0016] Specifically, the method for determining whether to remove or retain the turning point is as follows:

[0017] Given a threshold η∈[0,1] and a radius threshold R turn For the obtained set of turning points TP = {TP1, TP2, ..., TP} n}, traverse set TP, when TP i radius range R turn Number of low-speed points (TP) i .clp and TP i The number of all points within the radius TP i .cp satisfies TP i .clp>TP i When .cp×η, the current turning point is retained; otherwise, it is considered not to meet the conditions for a turning point and is discarded.

[0018] From the above description of the present application, compared with the prior art, the present application has the following beneficial effects:

[0019] The present application provides a kind of intersection extraction method based on low-speed feature and clustering algorithm, comprising the following steps: reading track data;Extract the turning pair in track, define two consecutive GPS positioning points as turning pair, and extract the corresponding turning point, obtain the turning point set;Filter out the low-speed point of all vehicle tracks in flat peak period;According to the number of low-speed points around each turning point, eliminate or retain the turning point, obtain the screened turning point dataset;Using clustering algorithm based on local connectivity, classify the screened turning point dataset to extract class cluster center, obtain intersection;The method proposed in the present application considers the low-speed feature of vehicle when passing through intersection, extracts turning point according to the change of vehicle heading, eliminates turning point according to the number of low-speed points, and obtains intersection information after clustering, with high accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flow chart of intersection extraction method based on low-speed feature and clustering algorithm is provided for the embodiment of the present application;

[0021] Figure 2 The turning point distribution area of intersection extraction algorithm based on low-speed feature and local connectivity clustering in the embodiment of the present application;

[0022] Figure 3 The distribution diagram of turning point and low-speed point of intersection extraction algorithm based on low-speed feature and local connectivity clustering in the embodiment of the present application;Fig. (a) is the extraction distribution of turning point, and Fig. (b) is the distribution of turning point and low-speed point;

[0023] Figure 4 The extraction result of intersection extraction algorithm based on low-speed feature and local connectivity clustering in the embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be described clearly and completely in combination with the drawings in the embodiment of the present application.

[0025] Embodiment:

[0026] Please refer to Figure 1 The present application provides an intersection extraction method based on low-speed feature and clustering algorithm, which comprises the following steps:

[0027] Step 1: reading track data.

[0028] Step 2: extract the turning pair in track, and extract the corresponding turning point.

[0029] Step 3: Filter out the low-speed points of all vehicle trajectories during off-peak hours.

[0030] Step 4: Eliminate or retain the turning point based on the number of low-speed points around it.

[0031] Step 5: Use a clustering algorithm based on local connectivity to classify and extract cluster centers for different intersections.

[0032] This embodiment uses GPS trajectory datasets of taxis and ride-hailing vehicles in Xiamen provided by a competition on the Xiamen Big Data Security Open Platform. The time period is the Dragon Boat Festival period in 2019 and 2020 (June 6-9, 2019 and June 24-27, 2020). The data dictionary is shown in Table 1.

[0033] Table 1 Data Dictionary

[0034]

[0035] The experimental area is located in Huli District, Xiamen City, with longitudes between [118.121926, 118.180926] and latitudes between [24.498138, 24.548138]. Figure 2 As shown, there are a total of 126 intersections, and the trajectory data is also filtered by latitude and longitude range.

[0036] In step 1, the data read in is defined as follows:

[0037] The definition of trajectory data is: the trajectory set Traj=<s1,s2,...,s n > represents a set of n trajectories. One of these is a GPS trajectory s. i ={p1,p2,...,p m Cid i} represents a continuous trajectory of points, where p represents a GPS point, and Cid... i This represents the vehicle number of the i-th trajectory. A consecutive trajectory is a set of location points with adjacent time intervals within 10 minutes.

[0038] GPS point p is described as p i (lgt,ltt,drc,v), where lgt represents longitude, ltt represents latitude, drc represents heading angle, which is the angle between the current driving direction and due north clockwise, and v represents the current vehicle speed.

[0039] In step 2, the steering pair is defined as follows: when a vehicle passes through an intersection and completes a turning maneuver, it leaves at least two consecutive adjacent trajectory points p1 and p2, where the heading angle of p1 is p1.drc, the heading angle of p2 is p2.drc, and Δα is the angle between the headings of the two points p1 and p2. A steering angle threshold α is given. turn and a time threshold t turn When Δα is greater than the threshold α turn And the time difference between the two points is less than the threshold t. turn At this point, we call p1 and p2 a turning pair, and the intersection of the extensions of the headings of trajectory points p1 and p2 is the turning point of this turning pair, denoted as TP. 1,2 .

[0040] p1(x1,y1,k1) and p2(x2,y2,k2) are a steering pair, where x, y, and k are the coordinates and slopes of the corresponding Cartesian coordinate system, respectively. The turning point TP(x,y) of the steering pair is calculated as follows:

[0041]

[0042] Traverse all trajectories, and when the difference in heading angle between two trajectory points is less than the turning angle threshold α... turn Furthermore, the time difference between the two trajectory points is less than the time threshold t. turn Then the two trajectory points are determined to be a turning pair, and the final set of turning pairs TP = {TP1, TP2, ..., TP3} is obtained. n}

[0043] In step 3, for each trajectory point, its velocity is calculated. If the velocity v of the trajectory point is less than a set threshold, it is marked as a low-speed point.

[0044] In step 4, a threshold η∈[0,1] and a radius threshold R are given. turn Using the turning point set TP = {TP1, TP2, ..., TP} obtained in step 2 n}, traverse set TP, when TP i radius range R turn Number of low-speed points (TP) i .clp and TP i The number of all points within the radius TP i .cp satisfies TP i .clp>TP i When .cp×η, we retain the current turning point; otherwise, we consider it to not meet the conditions for a turning point and discard it.

[0045] In step 5, for any two points TP in the set of turning points TP...i (ltt i ,lat i ) and TP j (ltt j ,lat j ), using

[0046] C = sin(lat i ) x sin(lat j ) x cos(ltt i -ltt j ) + cos(lat i ) x cos(lat j )

[0047] geoDistance(TP i ,TP j ) = R x arccos(C) x pi / 180

[0048] Calculate their Euclidean distance geoDistance(TP i ,TP j ), then use Algorithm 1 of Figure 2 to extract the intersection.

[0049] Evaluation method and index

[0050] In order to quantitatively compare the performance between different methods, this chapter uses two indexes of precision and recall to evaluate the effectiveness of map matching algorithm, whose formula is as follows, where the values of N tp , N fp , N np respectively represent the number of correctly extracted intersections, the number of incorrectly extracted intersections, and the number of unextracted intersections by the matching algorithm.

[0051]

[0052] Parameter setting

[0053] According to repeated experimental verification, the threshold value of the angle difference between the steering pair is selected as 60° among 45°, 50°, 55° and 60° as the best value. The threshold value of the time difference between the steering pair is selected as 20s among 15s, 20s and 25s as the final value. Through the visual analysis of the experimental results, it is judged that the experimental effect is best when the statistical low-speed point range radius of the turning point is taken as 30m and the proportion of low-speed points in the radius range of the turning point is taken as 55%. The specific parameter settings are shown in Table 2.

[0054] Table 2 Parameter configuration table

[0055]

[0056] To analyze and verify the effectiveness of the algorithm in this paper, comparative experiments are conducted between the algorithm in this paper and the following algorithms, and the experimental results are compared.

[0057] Algorithm 1: Intersection recognition based on local connectivity clustering algorithm (2017) uses the different turning sizes, different turning numbers, and different turning types of vehicles at intersections and non-intersections to filter turning pairs and turning points, and uses the local connectivity clustering algorithm to calculate the intersections;

[0058] Algorithm 2: Intersection extraction based on DBSCAN clustering algorithm of multi-density region division (2019) uses the different turning sizes, different turning numbers, and different turning types of vehicles at intersections and non-intersections to filter turning pairs and turning points, and uses the DBSCAN clustering algorithm of multi-density region division to extract intersections;

[0059] The experimental data selected by our experiment is the GPS trajectory data of cruising cars and online car-hailing vehicles in the time range from 10 am to 16 pm during the city's flat peak period. As shown in Figure 3 , the extraction results of the turning points and the distribution results of the low-speed points on the turning points are shown, where Fig. (a) is the extraction distribution of the turning points, and Fig. (b) is the distribution of the turning points and the low-speed points. The low-speed points during the city's flat peak period are almost included in the range of the turning points extracted by the algorithm. It can be clearly seen that the low-speed points of the vehicle trajectory during the city's flat peak period are more concentrated at the intersections than the turning points extracted by the turning pairs. By combining the two methods, we can more accurately extract the spatial position of the city's intersections. As shown in Figure 4 , the result map of the final clustering center based on the low-speed feature and the clustering algorithm is shown.

[0060] The low-speed feature of the city's flat peak period proposed in this paper is applied to the intersection extraction based on the local connectivity clustering algorithm and the intersection extraction based on the DBSCAN clustering algorithm of multi-density region division, respectively. Since the feature distribution of the low-speed GPS points during the city's flat peak period is considered for filtering the turning pairs, the traditional method of extracting the turning pairs is improved by considering the proportion of the low-speed GPS points within a certain range near each turning pair during the flat peak period. If the proportion reaches a certain threshold, it is retained, otherwise it is discarded, thus greatly improving the accuracy of the intersection extraction algorithm. The pseudo code of the intersection extraction algorithm based on the low-speed feature and the local connectivity clustering is shown in Table 3.

[0061] Table 3: Pseudo code of intersection extraction algorithm based on low-speed feature and local connectivity clustering

[0062]

[0063] The algorithm comparison results are shown in Table 4, and the experimental results show that after the intersection extraction algorithm with the low-speed feature is introduced, the precision and recall rate of the algorithm are obviously improved, wherein the precision of the DBSCAN clustering algorithm based on the low-speed feature and multi-density division reaches 97.54%.

[0064] Table 4 Algorithm comparison experimental results

[0065]

[0066]

[0067] The application provides an intersection extraction method based on a low-speed feature and a clustering algorithm, and the method comprises the following steps: reading in trajectory data; extracting a turning pair in the trajectory, defining two continuous GPS positioning points as the turning pair, and extracting a corresponding turning point to obtain a turning point set; screening out low-speed points of all vehicle trajectories in a flat peak period; according to the number of low-speed points around each turning point, the turning point is removed or retained to obtain a screened turning point data set; a clustering algorithm based on local connectivity is used to classify the screened turning point data set, extract a class cluster center, and obtain an intersection; the method provided by the application considers the low-speed feature of a vehicle when passing through an intersection, extracts a turning point according to a heading change of the vehicle, removes the turning point according to the number of low-speed points, and obtains intersection information after clustering, so that the precision is high.

[0068] The above merely describes specific embodiments of the application, but the design concept of the application is not limited to this, and any non-essential change to the application according to the concept should be regarded as an infringement of the protection scope of the application.

Claims

1. A method for intersection extraction based on low-speed features and clustering algorithms, characterized in that, Includes the following steps: Step 1: Read in trajectory data; Step 2: Extract the turning pairs from the trajectory. Define two consecutive GPS positioning points as a turning pair and extract the corresponding turning points to obtain the set of turning points. Step 3: Filter out the low-speed points of all vehicle trajectories during off-peak hours; Step 4: Based on the number of low-speed points around each turning point, remove or retain the turning point to obtain the filtered turning point dataset; Step 5: Use a clustering algorithm based on local connectivity to classify the filtered turning point dataset and extract cluster centers to obtain the intersections; In step 1, the data read in is defined as follows: the trajectory data is defined as a trajectory set. , represents a set of n trajectories, where one is a GPS trajectory. For a continuous trajectory of points, Represents a GPS point. Indicates the first The vehicle number of the track; a consecutive track point is a set of location points with adjacent time intervals within 10 minutes; GPS point descriptions are as follows. , Indicates longitude. Representing dimension, This indicates the heading angle, which is the angle between the current direction of travel and due north, rotated clockwise. Indicates the current speed of the vehicle; In step 2, a steering pair is defined as follows: when a vehicle passes through an intersection and completes a turning maneuver, it leaves at least two consecutive adjacent trajectory points. and ,in The heading angle is , The heading angle is ,and Two points and The angle of the heading; given a turning angle threshold and a time threshold ,when Greater than the threshold And the time difference between the two points is less than the threshold. At that time, it was called and For a pair of turning points, and trajectory points course and The intersection of the extensions of the headings is the turning point of this pair of turns, denoted as . ; and For a pair of steering pairs, where They are respectively The slopes of the x and y coordinates and their headings in the corresponding Cartesian coordinate system. They are respectively The slopes of the horizontal and vertical coordinates and their corresponding Cartesian coordinate systems; the turning points corresponding to the steering pairs. The calculation method is as follows: ； Iterate through all the trajectories, and when the difference in heading angle between two trajectory points is less than the turning angle threshold... Furthermore, the time difference between the two trajectory points is less than the time threshold. Then the two trajectory points are determined to be a turning pair, and the final set of turning pairs is obtained. ; In step 3, for each trajectory point, its velocity is calculated. If the velocity of the trajectory point... If the speed is less than the set threshold, it is marked as a low-speed point; In step 4, a threshold is given. and a radius threshold Using the set of turning points obtained in step 2 traverse the set ,when radius range Number of low-speed points and The number of all points within the radius satisfy If the current turning point is not met, it is retained; otherwise, it is considered that the turning point does not meet the conditions and is discarded. In step 5, for the set of turning points Any two points in and ,use ； Calculate their Euclidean distance Extract the intersection; To quantitatively compare the performance of different methods, accuracy is used. and recall rate Two metrics are used to evaluate the effectiveness of map matching algorithms, and their formulas are as follows: ； ； in, , , The values ​​represent the number of intersections correctly extracted by the matching algorithm, the number of intersections incorrectly extracted, and the number of intersections not extracted, respectively.

Images