Research On Abnormal Detection Method Based On Trajectory Data Mining

At present,the anomaly detection based on floating car GPS data has become a hot topic in the field of intelligent transportation and data mining.The goal of anomaly detection is to detect the anomalous behavior of the vehicle in real time and alarm in time,which is the key to realize the intelligent traffic monitoring.However,the uncertainty of data,the sparsity of the features and the huge amount of data have a great impact on the trajectory data mining methods and the quality of results.In this paper,the anomalous behavior mining of taxi trajectory data was studied deeply.Based on the data cleaning and processing,the different characteristics of the trajectory was fully utilized to propose the feasible trajectory and traffic anomaly detection methods from the micro level to the macro level.The main work of the paper is presented as follows.(1)According to the characteristics of taxi track data in Xi'an,the data were cleaned the data and the passenger trajectories were extracted.Through filtering,deleting the repeat value and other preprocessing work,the anomalous frequency data,repeat data or latitude and longitude range of abnormal data were screened and cleaned.The passenger trajectories were extracted from the filtered data based on the the vehicle's state value.Then combining with the existing road network information,the multi-weights map matching method was adopted for GPS points mapping based on the azimuth and distance considering location,time,direction and other information.The trajectories were matched accurately to the road.(2)This paper designed an expression method named intersection sequence track for trajectory based on the road network information,and used the trajectory anomaly detection method based on Bayesian principle to detect the intersection sequence track.This part first utilized the existing intersection data to express the trajectory in the form of a unique intersection sequence,which effectively reduced the data complexity.Then,calculating the probability of transition of each intersection based on the number of tracks passed through the intersection and Bayesian principle,the transition probability matrix of intersection was constructed.And then the path probability of the trajectory was calculated based on the transition probability matrix.Finally,comparing the path probability with the setting threshold to judge whether the trajectory was abnormal or not.The results showed that this method could detect the abnormal track effectively,and could accurately detect those abnormal trajectories,e.g.,abnormal line travel,detours like abnormal tracks or local abnormal tracks.(3)This paper proposed a method to construct the common path pattern between OD pairs based on space,time and driving behavior.Then an anomaly detection strategy of traffic flow caused by abnormal path change was proposed based on this pattern.In this part,the road network was divided into grids and expressed the track in the form of a grid sequence.Then,the trajectory anomaly detection standard was established from three aspects,space,time and driving behavior.The conventional path pattern between OD pairs was constructed on the data set excluding the abnormal trajectories.Finally,the traffic anomaly detection standard was established through the change of traffic flow caused by the change of the path pattern.The traffic anomaly caused by traffic accidents were tested and verified with the experiments and the results showed that the proposed method could detect traffic anomaly effectively.In this paper,the two-level anomaly detection methods were experimentally verified by real taxi GPS data in Xi'an,nearly 5000 tracks(60000 points)used in experiments.The computer configured Inter Xeon 2.50 GHz CPU and 16.0 GB memory,which could realize the anomaly detection within 3 minutes.Comparing with the traditional algorithms of the same level,the methods proposed in this paper had significant improvement in efficiency and accuracy,which met the needs of traffic management system for abnormal detection algorithm.