上下文感知的改进Staple跟踪算法

为解决相关滤波算法受余弦窗和搜索区域限制,导致跟踪器无法学习更多背景信息的问题,结合Context-Aware和Staple跟踪算法提出一种自适应特征融合的抗遮挡目标跟踪算法。在跟踪器中嵌入遮挡判断模块,利用平均峰值相关能量及峰值信噪比(PSNR)等指标来判断目标的遮挡情况,决定是否更新模板。将该算法在OTB-2015测试集上与其它7个先进的算法进行比较,实验结果表明,该算法的精确度与成功率分别为0.818和0.731,相对于Staple算法分别提升了4.3%和4.6%,跟踪速度达到34.55帧/s,验证了其有效性。

Improved Stable tracking algorithm based on context-aware

To solve the problem that the correlation filtering algorithm is limited by cosine window and search area,which makes the tracker unable to learn more background information,an anti-occlusion object tracking algorithm based on adaptive feature fusion was proposed by combining the Context-Aware and Staple tracking algorithm.By embedding the occlusion judgment module in the tracker and using the average peak correlation energy(APCE)and peak signal-to-noise ratio(PSNR)and other indicators,the occlusion of the target was judged,so as to decide whether to update the template.Compared with other seven advanced algorithms in OTB-2015 test set,experimental results show that the accuracy and success rate of the algorithm are 0.818 and 0.731 respectively,which are 4.3%and 4.6%higher than that of the Stable algorithm,and the tracking speed is 34.55 FPS,which verifies the effectiveness of the proposed algorithm.