自适应车流方向的跨道行人检测

摘要

针对视频交通事件监控中的行人检测问题,提出一种自适应车流方向的跨道行人检测算法.利用运动历史图像计算画面中物体的运动方向,通过学习获得运动参考方向场,并对运动目标进行跟踪.依据参考方向场分析其轨迹的走向,以确定运动目标是否发生跨道行为.实验结果表明,该算法能适应车流方向并有效检测跨道行人,具有较好的实时性和较高的检测成功率.%For video traffic events detection, a novel algorithm to detect crossing pedestrians is researched, which can be adaptive to the direction of traffic flows. It calculates the motion directions of objects in the video by Motion History Image(MHI) motion history images. A referenced motion direction field is obtained at learning step. When an object is tracked, its trajectory will be analyzed on line. Its motion direction is compared with the direction defined by the referenced motion direction field, which results in whether it is a crossing pedestrian. Experimental results show that this algorithm has the ability to adapt the traffic flow direction, and that it can run at real time with great accuracy.