中文摘要：

时空切片方法是一种有效的时空分析方法。然而,现有的算法仅讨论处理近似直线的简单轨迹情况,不能满足实际存在复杂曲线轨迹的处理要求。针对这一问题,本文提出一种实时的时空切片复杂轨迹分析方法来实现人体运动跟踪。首先在视频不同高度处提取水平时空切片,在各切片中分别使用高斯背景模型检测人体轨迹;然后拼合不同高度切片中的人体轨迹,在拼合后的轨迹上使用Hough变换计算人体对应的轨迹方程;最后根据轨迹方程和轨迹检测结果,确定人体的当前坐标、宽和高等信息。实验表明,相对于传统跟踪方法,该方法降低了跟踪的轨迹误差,满足实时性跟踪要求,且在人体短时遮挡情况下仍然能够实现有效的人体跟踪。

英文摘要：

Spatio-temporal slice is an efficient spatio-temporal analysis method. However, the literatures arc limited to simple straight trajectory processing, which cannot fulfill complex trajectory handling in real-world scenarios. To solve this problem, this paper presents a real-time spatio-temporal slice complex trajectory analysis algorithm to track human motions. First, horizontal slices at different heights are extracted from the surveillance videos. Gaussian background modeling is car- ried out to detect human trajectories in these slices. Then, after combining all trajectories from different slices, Hough transform is employed to compute the combined trajectory equations. Finally, humans＇ positions as well as widths and heights are obtained based on the equations and trajectory detection results. Compared with traditional tracking approaches, experimental results show that our approach reduces trajectory errors, satisfies real-time tracking requirement and has abili- ties to track humans under short-time occlusions.