基于多分辨率光流和多尺度角点的运动车辆跟踪

Moving Vehicle Tracking Based on Multi-resolution Optical Flow and Multi-scale Corner Detection

针对稀疏光流LK(Lucas-Kanade)算法不能稳定跟踪快速移动目标的局限性,提出了基于小波金字塔的多分辨率光流跟踪算法.算法基于多分辨率思想对原始稀疏光流进行了改进,从而实现了准确跟踪快速移动目标.在特征提取方面,提出了多尺度Harris角点检测方法,较好地解决了传统Harris方法的漏检和角点分布不均匀的缺陷,适合复杂交通场景中运动车辆特征提取.实验表明,当运动车辆旋转、移动以及摄像机变焦时,角点始终稳定可靠,并且跟踪算法能够快速、准确地匹配特征角点,实现了复杂交通场景下对运动车辆目标的实时稳定跟踪.

A multi-resolution optical flow tracking algorithm based on wavelet pyramid is proposed,since the LK（Lucas-Kanade） algorithm for sparse optical flow cannot steadily and rapidly track the moving objects.By virtue of the idea of multi-resolution tracking and relevant computation,the algorithm decomposes the displacement of the object within the wavelet pyramid to enable the displacement to meet the requirement of the LK algorithm for tracking accurately the rapidly moving objects.Extracting the feature of moving vehicles,multi-scale Harris corner detection is proposed to adapt to the complicated traffic situation,thus solving the problem that the conventional Harris corner detection which may neglect corner points and their non-uniform distribution.Experimental results show that in this way the corner points are always steady and reliable when a vehicle is steering and moving,or the camera is zooming in/out,and the tracking algorithm proposed can provide an accurate and real-time match for the feature points.