结合Hausdorff距离和最长公共子序列的轨迹分类

Trajectory Classification Based on Hausdorff Distance and Longest Common SubSequence

为了提高运动目标轨迹分类的准确性,该文综合考虑了轨迹的位置信息和方向信息,提出了一种结合Hausdorff距离和最长公共子序列(Longest Common SubSequence,LCSS)的轨迹分类算法。该算法首先采用改进的Hausdorff距离对轨迹的位置信息进行相似性测量,然后采用改进的LCSS算法对轨迹的方向信息进行相似性测量。与其他轨迹聚类算法不同,该算法融合了Hausdorff距离和LCSS两种算法的优点,提高了轨迹分类的准确性。此外,为了进一步降低计算复杂度,该文还实现了一种基于插值的保距变换算法和一种LCSS快速算法。实验结果表明,该轨迹分类算法可以明显提高轨迹的聚类准确率,聚类准确率可达到96%;基于插值的保距变换算法和LCSS快速算法可以很大程度上降低算法的计算复杂度,下降幅度最大可达到80%。该方法可以同时满足轨迹分类对精确度、实时性和鲁棒性的要求。

Considering the position and direction of trajectories of moving objects, a trajectory classification algorithm is proposed based on improved Hausdorff distance and Longest Common SubSequence （LCSS） to improve the trajectories classification. In this algorithm, the position similarity between trajectories is measured by the modified Hausdorff distances. And then the direction of the trajectories is distinguished by the modified LCSS distances. Comparing with other trajectory classification algorithms, the proposed algorithm compromises the merits of both Hausdorff distance and LCSS in trajectory classification and enhances the trajectory classification accuracy. Furthermore, to reduce the computational complexity of the similarity measure, a method of modified isometric transformation algorithm and an LCSS fast algorithm are realized. Experimental results show that the clustering accuracy of the proposed algorithm is greatly improved and the clustering accuracy rate can achieve 96% Meanwhile, the computational cost is greatly reduced by the modified isometric transformation algorithm and the LCSS fast algorithm, and the magnitude of the declines can reach to 80%. The proposed algorithm can satisfy the system requirements of higher precision, real time and robustness.