基于光流场估计的自适应Mean-Shift目标跟踪算法

针对Mean-Shift算法在目标跟踪中出现由于目标运动速度过快或尺度明显变化以及目标遮挡时导致跟踪失败的问题,结合光流场估计,提出了一种自适应Mean-Shift跟踪算法。本文方法在基于传统均值漂移矢量法的同时,引入光流法,在目标上找寻特征点,通过特征点前后变化的信息,修正跟踪窗口中心位置和大小,再根据Bhattacharyya系数二分法分别自适应得到更为精确的窗口长宽;而针对目标被静止物体遮挡,通过色差分析观测目标被遮挡区域,利用Bhatta-charyya系数重新捕捉目标。实验结果表明,本文方法在对目标移动方向较明显或由透视变化而导致的尺度变化具有较其他算法更优异的表现。将本文方法应用到铁轨跟踪实际中,测试结果表明,结合本文方法可显著提高轨道跟踪的可靠性。

An adaptive Mean-Shift algorithm based on optical-flow field estimation for object tracking

The Mean-Shift algorithm often fails when tracking a target with a high speed,a large change of scale or an occlusion.To tackle the problem,taking advantage of optical flow method,we propose an adaptive Mean-Shift object tracking algorithm in this paper.This method is based on the mean drift vector of the tracking window center,and Bhattacharyya coefficient based dichotomy is used to get both width and height of the window.Especially,the optical Flow method is employed to fine-tune the window position and window size according to the information of feature points.To track targets which are occluded by immobile objects,we use the color difference to observe the occlusion zone,and catch the target with Bhattacharyya coefficient when it is unsheltered.Experimental results show that the tracking algorithm has a very good effect in some cases.Besides,the proposed tracking algorithm has been successfully applied to rail tracking,and the application demonstrates that the algorithm can significantly improve the rail tracking reliability.