基于光流法与分块思想的目标跟踪算法

Object tracking algorithm based on optical flow method and block training

为提高目标在被遮挡、快速移动以及平面内旋转情况时的跟踪精度,提出基于光流法与分块思想的目标跟踪算法。基于光流法引入跟踪失败判别机制,将预测框中每个像素的光流值与整幅图片的平均光流值进行比较,依据超出阈值像素点数目的百分比对跟踪结果进行判别;若目标跟踪被判定为失败时,提出基于分块训练思想的模板更新机制改善后续跟踪过程,即把当前视频帧分割为多个子块,并单独训练每个子块获得对应的子滤波器模板,对各个子滤波器加权获得最终的滤波器模板。实验结果表明,该算法能够有效改善在目标被遮挡、目标快速移动以及目标平面内旋转问题下的跟踪效果。

For dealing with some difficult problems such as occlusion,motion blur and in-plane rotation in object tracking,the algorithm based on optical flow and blocking was introduced to achieve the purpose of improving the tracking accuracy.The tracking failure discrimination mechanism was proposed based on the optical flow method.The optical flow values of each pixel in the prediction frame were compared with the average optical current value of entire picture.The percentage of threshold exceeded was taken as the basis for the judgment.The filter template updating mechanism based on the idea of block training was proposed while the target tracking was judged to be a failure,which meant that the current video frame was divided and separately trained for sub-filter to improve the accuracy of tracking,and the final filter was got by weighting the sub-filters.Experimental results show that the proposed algorithm can effectively improve the tracking process under the problems like occlusion,motion blur and in-plane rotation.