CN115798212A

CN115798212A - Traffic jam detection method based on taxi track

Info

Publication number: CN115798212A
Application number: CN202211474284.9A
Authority: CN
Inventors: 何育枫; 林珲
Original assignee: Jiangxi Normal University
Current assignee: Jiangxi Normal University
Priority date: 2022-11-23
Filing date: 2022-11-23
Publication date: 2023-03-14
Anticipated expiration: 2042-11-23
Also published as: CN115798212B

Abstract

The invention discloses a traffic jam detection method based on taxi tracks, which comprises the steps of constructing a geographic scene of a taxi in a city, and taking an event-process as a center, namely an scene space-time dynamic model (EPCDM), a traffic jam detection model and a road grid speed evaluation model. The method comprises the steps that space connection operation is carried out on taxi tracks and a grid based on a traffic jam detection model, and whether the taxi in the grid is jammed or not is estimated and evaluated by estimating the number of stagnation points of taxis in the grid, namely GNSS stagnation points and the average speed; the invention abstractly establishes a traffic space-time dynamic model by taxi states, vehicle driving processes and traffic jam events, and realizes real-time traffic jam detection by the space-time data retrieval and reasoning capabilities of the dynamic model. The method can overcome the defects of expensive equipment, insufficient real-time performance and insufficient direction detection on the road congestion based on a machine learning method in the current traffic flow congestion detection method, quickly and accurately detect the traffic congestion and enrich the means of traffic congestion detection.

Description

Traffic jam detection method based on taxi track

Technical Field

The invention relates to the field of intelligent traffic, in particular to a traffic jam detection method based on taxi tracks.

Background

With the acceleration of urbanization pace, the problem of urban traffic congestion is more and more serious. Traffic congestion has become one of the most influential, longest lasting, and most frequently occurring of many traffic problems. To solve this problem, it is effective to quickly detect traffic congestion and then improve the congestion condition using knowledge of the cause of the congestion.

There are many detection algorithms for predicting traffic congestion. They can be classified into a conventional traffic flow detection method and a machine learning-based method. Conventional approaches have been directed to data characterization of traffic data within a particular range by detecting inherent traffic flow parameters and using physical or mathematical models to simulate traffic dynamics in other areas. First, conventional monitoring methods are limited by expensive monitoring equipment and the existence of blind spots when sampling data. Therefore, there is an error in estimating the traffic volume of unimportant roads and remote areas in a city. Secondly, road traffic flow estimation is based on analysis of a mathematical theory model, is weak in analyzing detailed vehicle behaviors, lacks of description on dynamic characteristics of traffic vehicle movement, and cannot explain the mechanism of congestion in a traffic process. Machine learning methods for traffic congestion detection and prediction are currently being developed rapidly. Traditional machine learning methods such as support vector machine SVM, classification trees, bayesian models, gaussian models, etc. are widely used. The neural network model applied to the field of traffic congestion judgment and prediction comprises a BP neural network, a Recurrent Neural Network (RNN), a multilayer feedback neural network, a RBF neural network, a Convolutional Neural Network (CNN), a Graph Neural Network (GNN) and the like. Compared with the traditional method, the machine learning algorithm can effectively detect traffic abnormity and can better identify the characteristics of traffic flow speed reduction, road occupancy rate increase, traffic jam and the like. However, existing machine learning lacks a uniform definition and measurement criteria for traffic congestion. The threshold for identifying congestion varies from region to region, and thus each criterion has no universality, which greatly increases the workload of identifying congestion. The deep learning method focuses on data feature mining and sample training, and converts the traffic jam detection and prediction problem into the classification and regression problem of the target to be detected through mining available features and complex high-dimensional operation. However, deep learning itself tends to be difficult to delineate how features of the network layer are selected, a black-box approach that lacks "interpretability. The method can accurately realize the detection and prediction of the traffic jam, but obviously neglects the concrete dynamic representation of the traffic jam process, is difficult to depict the travel process of the jam event, and cannot explain and explain the reason of the jam.

Disclosure of Invention

In a traditional GIS database model, how to effectively organize vehicle tracks from massive floating data and realize dynamic representation and behavior identification of each vehicle is a difficult problem, and it is needless to say that complex relationships such as vehicle-to-vehicle, vehicle-to-traffic jam, traffic accidents and the like are depicted in the traffic process. When the sampling rate is high enough, the taxi track contains a plurality of vehicle states, and fine-grained traffic conditions can be reflected. Therefore, the congestion detection can also be realized by quickly organizing and deducing the instant traffic state from the space-time track. The premise is to establish a dynamic representation model of massive space-time trajectories so as to support organization, retrieval and reasoning of massive space-time data. The geographic scene model enables an ordered organization of the geographic environment from geometry, spatial location, semantic descriptions, attribute features, evolution processes and relationships. As an extension of this work, an event and process-centric geographic scene dynamic model (EPCDM) is used to implement complex computations, dynamic simulations, and spatio-temporal reasoning for scene dynamic evolution. The invention provides an event and process-based urban traffic jam dynamic representation model, which constructs jam events, taxi processes, states and other related things and complex relationships (such as inclusion, development and nesting relationships) among the jam events, the taxi processes, the states and the other related things.

The model is adopted to organize taxi dynamic, explore and establish a historical dynamic model base of taxi tracks, and update the model through real-time track data. In the traffic detection method, the traffic jam is estimated by using the distribution rule of the GNSS points of the taxis in the grid unit in unit time and the average speed of the taxis in the grid. And (3) rapidly finding the route and the congestion direction of the floating vehicle in the congestion area by using a Cypher query language by utilizing the space-time reasoning and complex computing capability of the dynamic model and combining a Neosemantics plug-in of Neo4 j.

The invention provides a traffic jam detection method based on taxi tracks. The invention integrates a geographic scene data model, an event-process-centered space-time dynamic model (EPCDM), a real-time traffic jam detection process and a grid vehicle speed evaluation model.

Wherein,

the geographic scene data model is a geographic scene model which abstracts objects and relations in the urban congestion geographic environment into hierarchy nesting, integrity and correlation according to the geographic scene data model. The geographic scene is composed of: people, things, events, states, phenomena, and geographic processes. The taxi scene comprises the following components: the geographic event is a road jammed in a city, and the geographic process is a taxi driving process. Congestion events are typically composed of a number of slow or stationary vehicle travel events. The vehicle driving process is characterized by a temporally continuous chain of vehicle states (GNSS points). The attributes of the state are speed, passenger, location, direction, etc. According to the vehicle congestion semantics, the vehicle driving Process can be divided into congestion sub-Process processes _C (Congestion part) and Normal Driving sub-Process _N . In addition, there are sub-processes in traffic such as temporary stops, refueling/charging, waiting for passengers, drivers to rest, etc. To express the spatial-temporal uniformity, the spatial region is divided in the form of a grid. Things in an urban traffic scenario are typically referred to as gas stations, parking lots, schools, traffic lights, traffic signs, hospitals, CBDs, restaurants, residential quarters, etc., most of which can be crawled through POI points of interest of a network map service. The traffic jam event is composed of a jam center and a secondary jam area on the space. Road congestion directions can be divided into entering congestion routes and exiting congestion routes.

The event-process-centered spatio-temporal dynamic model is simulated by simulating a geographic scene or a geographic phenomenon in a discrete mode. The method is characterized in that the relations in the geographic scene are organized according to a scene structure relation, a scene interaction relation and an evolution development relation, and are organized according to a 'scene-sub-scene-process/event-state-scene data object' structure from top to bottom, so that the taxi scene is dynamically organized and simulated. Therefore, the main line of the taxi scene dynamic model organization is as follows: traffic jam event-vehicle driving process-taxi GNSS point state. The method comprises the following steps of:

(1) Taxi track node and relation extraction and storage

In a traffic congestion logic model, a developmental relationship refers to the sequence of geographic processes or states over time. The hierarchical relationship mainly refers to an event containing process, the process contains a state and the grid contains the state. The number of taxi GNSS points is enormous, calculated at 1 million taxis and a sampling rate of 10 s/time, yielding approximately 20 million floating point coordinates cumulatively every 5 minutes. In conjunction with the traffic congestion detection algorithm of the present invention, the data will be updated every 5 minutes. Historical data was also divided into different CSV file sequences at 5 minute intervals.

According to the organization method of the space-time dynamic model based on the graph database, two algorithms are respectively realized to organize taxi track data in traffic jam detection. First, process and state nodes in a scene are constructed. Then, containment between the process and the states, chronological relationships between the states (Next/Precede), relationships between the trellis and the states are created. Likewise, the relationship of grid owning POI can be further constructed. Attributes of taxi status are "ID, speed, X, Y, passenger, taxi, time".

(2) Congestion event related data object extraction and warehousing

And constructing a relation among the congestion event, the grid and the congestion route. And obtaining a taxi set in the congestion grid according to the 'included by' relationship between the congestion center grid screened out by the spatial join analysis result and the taxi state. Firstly, deducing the running track of each taxi in a congestion area, then extracting the congestion degree of grids around the central congestion area according to the running distance and time of each taxi, and finally further determining the secondary congestion range around the congestion center. By detecting the peak value grid of traffic jam, the jammed entering route and exiting route on the taxi track can be found.

The real-time traffic jam detection process comprises the steps of carrying out space link operation on massive taxi track data and a road grid, estimating whether jam conditions exist in the grid or not by estimating the number of taxi detention GNSS detention points in the grid and the average speed, realizing real-time update of traffic jam detection and a model, and keeping the real-time performance of the traffic jam detection. The general process is as follows: firstly, setting severe congestion parameters and detecting a traffic congestion center; and then detecting a secondary congestion area around the congestion center until the congestion disappears according to the congestion distribution characteristics, namely secondary congestion distributed to two sides of the road congestion center, and marking and drawing a congestion route. And finally organizing relevant information such as events, congestion routes, congestion directions, congestion positions and the like and storing the relevant information into a congestion event model.

The grid speed evaluation model is used for carrying out space link operation on a taxi track and a grid, and estimating the condition of whether the grid is congested or not by estimating the number of stagnation points of taxis in the grid, namely GNSS stagnation points, and the average speed. The main contents are as follows:

when the average speed of the interval is lower than 30 km/h, the traffic congestion is defined, and when the average speed of the interval is lower than 10 km/h, the traffic congestion is defined as severe traffic congestion, as shown in table 1.

TABLE 1 traffic Congestion levels and parameters

Since the major roads of a city are usually straight, most roads are smaller than the diagonal length of the grid. If there are no GNSS points in the grid cell, the speed of the grid cannot be identified from the taxi GNSS points. When there is only one GNSS data point, the vehicle speed must be greater than the speed of the vehicle, which will record two points extremely distributed along the diagonal, which corresponds to a traffic smoothing situation. Therefore, computing the number of GNSS points in the grid within 5 minutes can be used to detect the speed of the flow in the grid.

Assuming that h is the grid length, k is the time interval of GNSS point records, and c is the number of points in the taxi grid in the T period, the average speed of the taxi along the diagonal road of the grid is V

Assuming that the usual traffic light waiting time is 60 seconds, if the traffic flow is smooth,

is the count of a taxi GNSS point, wherein]Indicating rounding. The number of points c present in the grid for 5 minutes must be greater than

To filter out the effects of traffic signals:

during congestion detection, most roads are shorter than the diagonal of the grid, so the average speed of V and the grid can be calculated

Making a comparison to identify congestion, wherein

Indicating a congested grid.

The number of taxis in the grid is represented by "count", and the number of GNSS data points in the grid for any taxi i is C _i The sum of all taxis in the grid is counted over 5 minutes, since the road is shorter than the diagonal, the average speed in the grid

Is composed of

For traffic congestion level S, the maximum speed in the grid is V _S By using C in equation (1) _S Instead of C, find the count of points in the grid that meet this definition of congestion level, resulting in

The condition for detecting the congestion event R is

To detect the average speed along the road, the size of the grid needs to be appropriate. Many studies have used grids that are greater than 500 meters wide, meaning that they can only identify traffic flow in urban grids, but not the speed of vehicles on the road.

TABLE 2 detection parameters for different congestion levels

Assuming a sampling rate k of 10 seconds per point and a diagonal of 250m of the grid, the width is 180m. Then, vs and C _S The method realizes the evaluation of the average speed of the grid cells in the model once every 5 minutes, and the calculation result is the key for realizing the model and can be used as an index of traffic jam.

Compared with the prior art, the invention adopting the technical scheme has the following technical effects:

the method and the system adopt the taxi track of the city to detect the urban congestion, have the characteristics of high detection speed, wide range, low cost and the like, have the capability of detecting the specific direction and position of the traffic congestion, and can support the discovery of the details of the occurrence of the traffic congestion. From the perspective of a traffic management department, the method saves the time cost, the labor cost and the equipment cost spent on traffic jam detection, has the capability of backtracking of historical jam information and behavior analysis, and can further provide precious opinions on urban traffic transformation and planning.

Drawings

FIG. 1 is an overall block diagram of a method for traffic congestion detection based on taxi tracks;

FIG. 2 is a city taxi scene element and its relationship;

FIG. 3 is a detection flow diagram;

fig. 4 is a database establishment and update process.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments, but the present invention is not limited thereto.

As shown in fig. 1, the general technical route of the traffic congestion detection method based on taxi tracks of the present invention includes a taxi geographical scene model, a time-space dynamic model with an event-process as a center, and a traffic congestion detection model.

The taxi geographic scene model is constructed by abstracting objects in a taxi congestion geographic environment and organizing the relationship among the objects. The abstract scene objects of the present invention are shown in table 1, and the relationship between the objects is shown in fig. 2.

TABLE 1 urban taxi scene composition

The event-process-centered space-time dynamic model refers to the construction and data filling of a space-time dynamic model in a taxi scene;

as shown in Table 2, the spatio-temporal dynamic model includes three categories of hierarchical relationships, developmental relationships, and interrelationships. The variation of the relationship to the spatio-temporal evolution of all scene objects in table 1 is covered.

TABLE 2 relation of urban taxi scene objects

The traffic jam detection model is used for carrying out space link operation on massive taxi track data and a road grid, estimating and evaluating whether the grid is jammed or not by estimating the number of taxi stagnation GNSS stagnation points in the grid and the average speed, realizing real-time update of traffic jam detection and the model and keeping the real-time performance of traffic jam detection. The detection flow of traffic jam is shown in fig. 3, and the general process is as follows: firstly, setting severe congestion parameters and detecting a traffic congestion center; and then detecting a secondary congestion area around the congestion center until the congestion disappears according to the congestion distribution characteristics, namely secondary congestion distributed to two sides of the road congestion center, and marking and drawing a congestion route.

Considering reasonable computational consumption and time resolution of traffic dynamics, the trajectory data will be divided into a plurality of trajectory files at 5 minute intervals. The identification of urban traffic flow mainly identifies the traffic flow of urban main roads. The number of vehicles left in the grid and the average speed of the taxis reflect the average traffic situation over 5 minutes. The detailed steps are as follows: (1) Firstly, performing Spatial Join operation between a grid sequence and a track point map layer of a taxi for 5 minutes, and counting an average speed and a maximum retention GNSS point in each grid according to a relation between a grid object and a state related in an established dynamic model; (2) Then, corresponding V is set according to the congestion degree _s And C _s A value (see table 2 for details), calculating the speed of the grid according to the evaluation method of the grid, and identifying the congestion condition of the grid; (3) Taking the detection of a severe congestion area as an example, the distribution of severe congestion grids (the distribution of one grid or a plurality of continuous grids on a road) is judged first; (4) Determining a congested track route according to a vehicle track of a congested area; (5) And finally, setting secondary congestion parameters along the road, and extracting a secondary congestion grid to form a multi-level congestion structure.

The database establishment and update strategy is shown in fig. 4. The method can meet the requirement of constructing the database for historical data, and can keep the real-time performance of the database according to the new arrival track. Through the time-space integration of 'event-process-state-grid', the historical data model database can inquire, retrieve and reason the behavior characteristics of historical vehicles, deduces a core region and a secondary congestion region of vehicle congestion according to the running route of the vehicles in the congestion region, and obtains the detailed condition of the traffic congestion event.

The above-described embodiments are intended to be illustrative, and not restrictive, of the invention, and all changes that come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Claims

1. A traffic jam detection method based on taxi tracks is characterized by comprising the following steps: the method comprises the steps of constructing a geographic scene of the urban taxi, taking an event-process as a center, and then constructing a scene space-time dynamic model (EPCDM), a road grid speed evaluation model and a traffic jam detection model;

the urban taxi geographic scene construction provides an organization method of taxi space-time dynamic data based on a scene-event-process-state-object structure according to a geographic scene data model theory; integrating scene level nesting, space-time evolution and mutual relations into a whole, and constructing a scene space-time dynamic model taking an event-process as a center in a direct relation connection mode;

the process of constructing the scene space-time dynamic model with the event-process as the center is as follows: (1) Analyzing scene elements and relations of states, driving tracks and events of taxis, and realizing a space-time dynamic data model for organizing urban taxi scenes according to an EPCDM model; (2) Extracting scene data nodes and scene relations from the track data, developing an algorithm of storing the scene data nodes and relations according to the requirements of the congestion data model, realizing the database building of a historical track spatiotemporal data model, updating new track data to the spatiotemporal track model, and updating the detected congestion event to the congestion model database;

the core calculation formula of the grid speed evaluation model is as follows:

the number of taxis in the grid is represented by count, k is the time interval of GNSS points, and the number of GNSS data points in the grid of any taxi i is C _i Counting the sum of all taxis in the grid within 5 minutes, the average speed in the grid

Is composed of

Wherein h is the grid width; for traffic congestion level S, the maximum speed in the grid is V _S By using C _S To represent the count of points in the grid that meet this congestion level definition, to derive

The condition for detecting the congestion event R is

Vs and C _S T is traffic light waiting time;

the traffic jam detection model is used for carrying out space link operation on a taxi track and a grid, and estimating whether the taxi in the grid is jammed or not by estimating the number of stagnation points of the taxi in the grid and the average speed; considering reasonable calculation consumption and time resolution of traffic dynamics, the track data is divided into a plurality of track files according to the interval time of 5 minutes; the detection process comprises the following steps: firstly, setting severe congestion parameters and detecting a traffic congestion center; and then detecting a secondary congestion area around the congestion center until the congestion disappears according to the congestion distribution characteristics, namely secondary congestion distributed to two sides of the road congestion center, and marking and drawing a congestion route.

2. The traffic jam detection method based on taxi tracks as claimed in claim 1, wherein: the traffic jam detection model detection process specifically comprises the following steps: first in a grid sequence and rentPerforming Spatial Join operation between the trace point layers of the vehicle within 5 minutes, and counting the average speed and the maximum retention GNSS point in each grid according to the relation between grid objects and states related in the established dynamic model; then, corresponding V is set according to the congestion degree _s And C _s Calculating the speed of the grid according to the evaluation method of the grid, and then identifying the congestion condition of the grid; taking the detection of a severe congestion area as an example, firstly, judging the distribution of severe congestion grids; then, according to the vehicle track of the congestion area, determining a congestion track route; and finally, setting secondary congestion parameters along the road, and extracting a secondary congestion grid to form a multi-level congestion structure.