| With the rapid development of China’s economy and society and the rapid advancement of urbanization,residents’ demand for transportation is increasing day by day,which also causes traffic congestion,environmental pollution and resource depletion and other traffic problems.At the same time,the trajectory data generated by moving objects contains rich spatial and temporal information,which is of great help to the analysis and mining of the basic traffic laws and characteristics of residents’ travel,and provides important data support for urban traffic demand,traffic supply and traffic system management.This paper on the research progress of the study on the existing track data analysis,due to data sources,acquisition methods,data quality and other factors,there are some limitations in the trajectory study: first,the work is mostly based on one type of traffic data analysis of conventional mobile pattern,such as digging the individual travel mode of private cars.However,single data source cannot reflect the characteristics of vehicle movement behavior of different types.Secondly,the collected vehicle tracks may be sparse,which brings challenges to the analysis of track behavior characteristics.Based on the above status quo and the demand of data-driven traffic analysis,the following work is completed in this paper :(1)Preprocessing operation of sparse tracks is realized.Firstly,data cleaning was carried out to remove redundant and drift track data,and data was filtered based on urban geographic location and speed limit regulations.Then the map matching method for sparse trajectories is used to map the coordinates of sampling trajectories to road network space.Finally,basic characteristic parameters such as two-point time difference and acceleration are extracted to describe the trajectory information.(2)Based on the existing experimental data,a novel trajectory segmentation problem is defined.Then the vehicle behaviors are divided into five categories according to vehicle types and trajectory travel segments,namely,no-load and travel of taxi,no-load and travel of online taxi,and travel of private car.From the perspective of empirical analysis,the behavioral characteristics of travel time,travel distance and average speed are studied.(3)Vehicle behavior recognition is essentially a trajectory classification problem.In this paper,a vehicle behavior recognition framework based on the LSTM model with multi-attribute fusion of sparse trajectory is proposed.By embedding features such as travel distance,POI and weather,implicit information of vehicle sparse trajectory is captured and identified.Finally,this paper uses the sparse trajectory data sets of three types of vehicles to verify the above research methods.Firstly,the spatiotemporal distribution characteristics of travel behavior were statistically analyzed and visualized,and the origin and the destination points of different vehicles were clustered to find certain differences in the performance of travel volume in different regions.Based on the basic and semantic features of trajectory,the multi-attribute fusion LSTM model is used to mine and identify sparse trajectory information.Micro F1 and Macro F1 were selected as the evaluation indexes of the model,and the results show that the proposed method has better performance than the classification models such as XGBoost. |
