Research And Implementation Of Visual Analysis For Anomalous Trajectory Detection On Taxi Data

In recent years,with the proliferation of GPS devices,more and more trajectory data in the scene has been mined.Taxi is an important part of urban rail transit system.Anomalous traj ectory detection is a basic research point of trajectory data mining.It has important guiding significance for trajectory data cleaning and the discovery of anomalous events in urban traffic.This paper studies the research and implementation of visual analysis system for anomalous trajectory detection on taxi data.Through the study of taxi trajectory in the city,we find the pattern of travel traj ectory in the city,and then propose a feasible solution to the anomalous trajectory detection problem.To make anomaly trajectory detection perceptible,we build a visual analysis system to assist the researchers.First of all,this paper analyzes some problems existing in the traditional anomalous trajectory detection algorithm.Trajectory detection is a widely studied problem.The traditional anomalous traj ectory detection methods mainly focus on the trajectory data set of a given starting point.These methods only consider the trajectory to a certain extent without the characteristics of the combined road network and the commonality between different trajectory data sets.Aiming at the above problems and deficiencies in the process of anomalous trajectory detection,this paper proposes the Anomalous Trajectory Detection using Recurrent Neural Network(ATD-RNN).ATD-RNN is inspired by the learning and representation learning which is prominent in deep learning.Through the representation of the learned trajectory,the distance between different trajectories can be measured in a low-dimensional space,so that the classifier of the abnormal traj ectory can be trained based on the idea of density clustering or using a machine learning method such as a neural network.In order to learn the sequence information between the traj ectories well,and improve the the performance of anomalous trajectory detection,in this paper,we study the method Anomalous Trajectory Detection using recurrent neural network based on attention mechanism(ATD-att).The validity of the anomalous trajectory detection using recurrent neural network is proved by the verification of the anomaly detection in the taxi traj ectory data.In addition,this paper has carried out a sufficient experimental analysis of the robustness of the proposed model.Based on the proposed anomaly trajectory detection model ATD-RNN and ATD-att,this paper designs and implements a taxi trajectory anomalous detection visual analysis system.Through the research on anomalous trajectory detection,this paper proposes a set of flow for visual analysis of ATD,which can help researchers perceive the problems existing in trajectory anomaly detection,and then assist the researcher to train the anomalous trajectory detection model to improve the performances.We verify the effectiveness of this system through a demonstation on real world dataset.