Research And Implementation Of Trajectory Segmentation Method For Trajectory Big Data

Moving objects will generate massive trajectory data.At the same time,these trajectory data contain spatial attributes and temporal attributes,through which we can describe the motion behavior of a specific object in a certain time range.In the mass trajectories generated by moving objects,the trajectories collected can not be used directly due to the noise of position collector and the existence of object occlusion.Therefore,complex preprocessing is needed for trajectory.The scale of trajectory data is increasing,which leads to the increasing length of trajectories.The purpose of trajectory segmentation is to transform long trajectory into short trajectory by reasonable segmentation.Trajectory segmentation not only reduces the complexity of subsequent calculation,but also provides more semantic knowledge for trajectory.Existing trajectory noise filtering methods usually can only filter a single noise,and the filtering effect is not good when continuous noise occurs.The existing trajectory segmentation methods are usually for specific applications,and the migration is poor.Therefore,this paper has carried out detailed research from two aspects: track anomaly filtering and trajectory segmentation.Design and implement the trajectory outlier filtering algorithm and trajectory segmentation algorithm.There are three aspects as follows:(1)Combined with the spatio-temporal properties of trajectory data,we propose a method to filter outliers based on finding stable sub trajectory segments.First of all,this method searches for the stable sub trajectories in the trajectory based on velocity and angle changes.Then,the trajectories of similar trajectories and sub trajectories can be found from the stable sub trajectories to the adjacent sides.Finally,all the stable sub trajectories are merged to complete the filtering of the trajectory.(2)This paper presents a reactive unsupervised trajectory segmentation method(REGRASP).This method takes the attributes of trajectories as characteristics,and defines the loss function that measures the internal homogeneity of trajectory segments and the characteristic distance between adjacent segments.The process of trajectory segmentation based on this method is as follows: firstly,the trajectory segments are added randomly and repeatedly,and the trajectory segments are extended to the minimum length on both sides according to the loss function,then the unassigned points between segments are traversed and divided,finally,all segments search the best segmentation points locally in the adjacent segments.Experimental results show that,compared with the existing unsupervised segmentation methods,the proposed trajectory segmentation method can adjust the key parameters,speed up the iterative process through filtering strategy and early stop strategy,and judge whether to continue to search for the better solution by itself.It does not need prior knowledge to set the threshold,and improves the operation efficiency of the algorithm.In terms of segmentation results,this algorithm has more advantages than the other two algorithms in restoring the original annotation trajectory,but at the same time,the purity of the trajectory segment is slightly lower than the other two algorithms.(3)We have designed and implemented a prototype system for trajectory data management in big data environment.The system uses Map Reduce under Hadoop framework as the computing engine,and HBase as the database to store trajectory data.The system achieve the management of trajectory data and the visualization of query trajectory results.