采用高效卷积算子的长期目标追踪算法

Long-term Tracking Based on Efficient Convolution Operator

针对目标跟踪中存在的对严重遮挡和出视野的目标跟踪失败的问题,提出了一种基于高效卷积算子的长期目标追踪算法.该算法首先利用滤波器的检测得分计算当前目标是否受到遮挡,停止对遮挡目标尺度模型的更新;然后利用遮挡情况和外观模型置信度判断目标的可靠性,对不可靠目标的模型不计入外观模型更新序列中.当目标长期遮挡或者丢失时,利用融合空间权重、目标检测得分和最佳伙伴相似性得分来重定位目标.与原始的高效卷积算子算法进行对比实验的结果表明,改进的算法能有效地解决目标遮挡和出视野情况下的目标跟踪失败的问题,具有较高的跟踪精度和鲁棒性.

Aiming at the problem of failing to track the severely occluded and out of sight target,a long-term tracking algorithm based on Efficient Convolution Operators for Tracking is proposed. First,the score of filter detection is used to calculate whether the current target is occluded and stop updating the occlusion target scale model in this algorithm. Then,the occlusion and the confidence of appearance model are used to judge the reliability of the target. When the target is occluded or lost for a long time,the fusion space weight,detection scores of the target and the best partner similarity score are used to reposition the target position. Compared with the original Efficient Convolution Operators for Tracking,the experimental results show that the improved algorithm can solve the problem of target tracking failure in the case of occlusion and out of sight. This algorithm has high tracking accuracy and robustness.