基于多尺度区域划分的车辆轨迹预测

In this paper, the vehicle trajectory prediction at regional level is studied, and a prediction method based on multi-scale regional division is proposed. Firstly, a multi-scale region division algorithm based on the origin-destination (OD) distribution of trajectories is proposed. The algorithm uses a two-phase strategy to divide the area to grids and merge neighbouring grids into sub-regions. The algorithm can adjust the size and shape of the sub-area according to the distribution characteristics of OD. Compared with the equal-scale meshing, the proposed algorithm avoids the destruction of the edge of the hot zoneand distinguishes the dense area and the sparse area. Secondly, after area dividion, a multi-order Markov model based on hash table is proposed. Compared with the traditional implement based on transition probability matrix, the hash table based implement effectively alleviates the problem of state space expansion and supports incremental updating. In this paper, the performance of the algorithm is verified based on the actual trajectory data. Experimental results showthat the proposed area division algorithm greatly improves the accuracy of trajectory prediction.The influence of the training data scale and the model orderon the prediction algorithm are analyzed.