基于矢量量化和深度优先搜索的轨迹分布模式学习算法

A novel trajectory pattern learning method based on vector quantization and depth first search

在对轨迹流矢量进行量化编码的基础上,提出了一种基于深度优先搜索的轨迹分布模式提取算法,生成了能够描述轨迹分布的序列模式图,并给出了与之相应的异常检测方法。对不同场景的序列图像的实验表明,该方法不仅能够学习轨迹中流矢量的分布,而且能够反映它们之间的时序关系,可以应用于室外复杂场景的目标异常行为检测。

This paper puts forward a novel depth first search method to learning the distribution of motion trajectory based on the vector quantization of the flow vectors. Then the sequential patterns graph is generated, which visually represents the trajectory pattern. The corresponding anomaly detection method is also given in this paper. The experiments on different sites demonstrate that our method can not only discover the flow vector＇s distribution but also can reflect their time orders effectively, which makes it suitable for anomaly detection in outdoor scenes.