基于二阶非线性投票的多目标跟踪算法

Multiple objects tracking based on two-stage nonlinear voting

提出了基于二阶非线性投票的多目标跟踪算法。该算法通过目标匹配得到同一目标在不同帧中的位置,同时利用特征监测来处理目标的遮挡、分裂问题,并实现目标特征的实时更新。在目标匹配过程中,通过对目标前一帧与当前帧的特征相似性进行投票,得到匹配目标。利用视频图像进行实验,结果表明,该方法对噪声、阴影、遮挡、分裂等具有良好的鲁棒性,较好地实现了多目标的跟踪。

This paper proposed a two-stage nonlinear voting-based tracking method. The method used object matching to get objects＇ position in different frames, and used feature monitoring to deal with object occlusion, object split and real-time update for objects features. Implementecl objects matching based on the similarity voting of their features in successive frames. Carried out experimental study using image sequences captured in real scene. The experimental results show that the method is robust against noise, shadows, occlusion and split, and it performs multiple objects tracking finely.