Research On Trajectory Similarity Based On Vehicle Behavior

On the premise of the increasing traffic big data,the calculation of vehicle trajectory similarity has become a hot spot in the research of traffic big data.Vehicle behavior refers to the trajectory selection of vehicles under different traffic conditions,which is the main cause of the difference of trajectory similarity.At present,the traditional single trajectory similarity calculation method mainly analyzes whether the two trajectories are similar in spatial position,while the traditional multi trajectory similarity calculation method has some problems,such as the combination of vehicle OD features and trajectory features is not close.Under the background of traffic big data project and referring to the research and judgment opinions of traffic police,this thesis studies the similarity calculation methods of vehicle single trajectory and multi trajectory from the perspective of vehicle behavior.The main research work and innovations are as follows:(1)In order to highlight the difference of vehicle behavior in the calculation method of vehicle single trajectory similarity,this thesis analyzes a similar vehicle behavior and two different vehicle behaviors,including short-term detour for the same purpose,long-term detour for different purposes and long-term parallel.Based on the features of vehicle behavior,the editing distance algorithm is improved,and a single trajectory similarity calculation method based on continuous trajectory editing length is proposed(CTEL).Compared with DTW,LCSS and ED methods,CTEL has stronger trajectory discrimination and meets the actual research and judgment needs of traffic police.(2)In order to improve the evaluation effect of vehicle multi trajectory similarity,combined with vehicle behavior,a multi trajectory similarity calculation method based on trajectory graph is proposed in this thesis.In this method,the vehicle multi trajectory is transformed into the vehicle trajectory graph,the OD feature sampling weight parameters are designed,and a random walk scheme of the trajectory graph with rich vehicle trajectory features is realized.The vehicle multi trajectory similarity is obtained through Word2 Vec model.In the model accuracy comparison experiment,the average accuracy of the Top10 of the bus is83.5%,and the average accuracy of the Top100 of the taxi is 94.3%.(3)Based on the similarity calculation method of trajectory graph,this thesis designs an application scheme of similar vehicle set construction and an application scheme of suspected operating vehicle discovery.Compared with the traditional methods of traffic police,the former has a higher recall rate,while the latter can screen non taxis with high similarity to taxis as suspected operating vehicles,providing basis for further screening and judgment of traffic police.There are 63 figures,18 tables and 84 references in this thesis.