Research On Algorithms For Detecting Transportation Modes Considering Stay Point Features

Transportation modes detection based on spatial and temporal trajectories is an important part of spatial-temporal trajectory mining.It aims to identify people's transportation modes,such as walking,taking a bus and so on,through their trajectories.Transportation modes detection has great value on travel recommendation,personalized advertisement recommendation,track retrieval,healthy life planning and so on.Most methods of transportation modes detection use moving features extracted from trajectories to help detect transportation modes.However these algorithms ignore the important location information(such as bus station)in trajectories.In order to improve the shortcomings of existing algorithms,this paper improves the effectiveness and efficiency of the transportation modes detection algorithm.The main innovative contributions of the thesis are achieved as follows:1.An algorithm for transportation modes detection considering stay points features is proposed.The algorithm takes into account the effect of the important position information in trajectories on transportation modes detection.The algorithm is divided into two stages,training stage and testing stage.In training stage,moving features of training trajectories arc extracted firstly.Then stay points in training trajectories are extracted.Stay points clustering rules and path rules are mined.Based on these rules,stay points features are produced for each trajectory.Finally,a classifier is established by training trajectories features containing moving features and stay points features.In testing stage,we first extract moving features of test trajectories,then extract stay points features by stay points clustering rules and path rules.Finally transportation modes are detected by the classifier.The experiment result shows that the algorithm can achieve good detection accuracy.2.An online algorithm for transportation modes detection considering stay points features is presented.The algorithm is divided into three parts:off-line training,online training and testing.Off-line training stage and testing stage are similar to the algorithm mentioned above.In online training stage,the original stay points clustering rules,path rules and classifier are updated based on the new training trajectories to have stronger abilities to detect transportation modes.The algorithm updates classification model in real time with the new coming training trajectories to improve the recognition accuracy.The experiment result shows that the algorithm has better scalability compared with off-line algorithm.3.Parallel transportation modes detection algorithms for the two algorithms mentioned above are proposed and the parallel algorithms are implemented by using the Spark platform.Computing tasks are distributed to multiple machines in parallel,and the operation efficiency of the algorithms are improved when the number of trajectories is large.The experiments results show that the efficiency of the parallel algorithms are higher than the single algorithms as data set increases.With increasing computing nodes,the algorithms proposed have good augmentability.