基于背景感知的相关滤波抗遮挡优化算法

Correlation filter anti-occlusion optimization algorithm based on background-aware

针对相关滤波目标跟踪中由于目标遮挡、出视野引起的跟踪失败等问题,以背景感知相关滤波(background aware correlation filters,BACF)算法为基础,提出一种抗遮挡优化算法。首先特征提取部分融合方向梯度直方图(histogram of oriented gradient,HOG)特征与颜色名称(color name,CN)特征;其次提出层级遮挡优化算法,在目标跟踪过程中根据平均峰值相关能量(average peak to correlation energy,APCE)和最大响应峰值判断目标外观是否发生较大变化,并通过巴氏距离进一步计算目标相邻滤波器模板特征相似程度,以此判断造成目标外观变化的真正原因,进而决定是否更新模板;针对目标出视野问题,提出模板筛查策略,跟踪过程中每隔K帧对滤波器模板进行置信度考查,若目标丢失,则生成新的模板并对目标进行全局搜索重新抓取目标。通过在目标跟踪数据集OTB2015上测试,优化算法精确度与成功率分别为83.0%与78.8%,有效提高了算法性能。

Aiming at the problem of object tracking failure caused by occlusion and out of view in correlation filter tracking,we propose an anti-occlusion optimization algorithm based on background aware correlation filters(BACF).First,the feature of histogram of oriented gradient(HOG) and color name(CN) are fused in the feature extraction part;Secondly,a cascaded occlusion optimization algorithm is proposed to determine whether the target appearance has changed significantly during the object tracking process based on the average peak correlation energy(APCE) and the maximum response peak,and further calculate the similarity of features between the target neighboring filter templates by the Bhattacharyya distance to determine the real cause of the target appearance change,and then decides whether to update the template;For the problem of target out of view,a template inspection strategy is proposed and the confidence of the filter template is checked every K frames during the tracking process,and if the target is lost,a new template is generated and the target is recaptured by a global search.By testing on the object tracking dataset OTB2015,the accuracy and success rate of the optimized algorithm are 83.0% and 78.8%,respectively,which effectively improve the performance of the algorithm.