鲁棒的实时多车辆检测与跟踪系统设计

A Robust System for Real-time Multiple-Vehicle Detection and Tracking

场景中的运动阴影导致多目标粘连,车辆间的相互遮挡使得跟踪识别困难。本文针对这两个影响实时车辆检测与跟踪系统性能的主要因素,采用基于无偏卡尔曼滤波器(UKF)的方法为场景背景建模,提取出运动区域,再通过边缘特征检测出场景中的运动阴影,然后利用角点信息将目标与阴影分离;提出了一种基于运动预测框的目标跟踪算法,将它与基于车辆平行四边形轮廓的遮挡分割方法结合,构建了多车辆目标的实时跟踪系统,并用实验验证了它的实用性与鲁棒性。

Moving objects in the video images may be connected by moving shadows, and vehicles occlusion would bring trouble to tracking and recognition. In order to solve these problems in the construction of a robust vehicle tracking and recognition system, an unscented Kalman filter （UKF） based method is adopted to model the scene background and extract the moving areas. Then the moving shadow areas are detected through edge feature and separated from the objects using corner point information. Movement estimation frame based tracking method is proposed, and combined with parallelogram contour based occlusion segmentation method to realize realtime multi-vehicle tracking. Experimental results demonstrated its practicality and robustness.