Research On Pattern Accumulation Trajectory Compression Method Under Road Network Constraints

With the advance of hardware devices and mobile computing technology,vehicular positioning systems,smartphones,and other mobile devices with positioning capabilities have been widely applied in people’s lives,leading to a rapid increase of trajectory data.In order to facilitate statistical analysis based on road networks and reduce storage consumption,online road network matching and trajectory compression are key factors in enhancing the application based on trajectories.However,current online map-matching algorithms do not fully make use of the temporal dependencies of trajectory points,resulting in low matching accuracy.Further,existing online trajectory compression methods failed to fully exploit the regularity of urban vehicle trajectories.To address these issues,we propose an online road-network trajectory compression method based on pattern accumulation,which focuses on the following two issues:1.During the preprocessing of road network trajectory,a window-based online map-matching algorithm is proposed by taking advantage of Hidden Markov Model(HMM).This algorithm introduces segmentation points,where if the candidate points of the road segment to be matched have similar probabilities,the streaming data flow is segmented at that point to form a matching window.When subsequent trajectory point arrives,the probabilities are reinitialized.Each window uses an independent HMM to choose the optimal matching path.Additionally,we propose to consider the distance,direction,and lane factors to calculate the emission and transition probabilities more accurately.2.To reduce the storage cost of urban vehicle trajectories,a trajectory compression method is proposed based on online pattern accumulation.This method represents a trajectory as a sequence of patterns,where each pattern represents a consecutive sequence of road segments that repeatedly appear on the road network.The compression is divided into three stages: In the first stage,the trajectory is transformed into a series of spatial component sequences and velocity component sequences based on the difference between the average velocity on the sequence of road segments and the average velocity on any subsequence starting from an anchor position.In the second stage,unique pattern sequences are extracted from the spatial component sequences,and their corresponding velocity component sequences are stored directly.In the third stage,each unique pattern sequence is represented by an identifier and stored in a pattern dictionary,so that each trajectory can be represented as a series of pattern identifiers.