【摘要】 针对机器视觉技术在嵌入式平台的应用,提出了具有遮挡检测和自适应模板更新的核相关滤波器跟踪模型,旨在提高跟踪器在嵌入式平台上运行时对尺度变化和遮挡的鲁棒性。为了应对目标运动过程中目标尺度变化,在核相关滤波器中加入多尺度估计与动态模板更新策略,同时为了保证目标遮挡后能被再次检测,采用粒子滤波器对多个粒子候选者重采样来检测目标。实验表明,跟踪器对OTB数据集测试的平均距离精度(mean DP)和平均重叠率精度(mean OP)分别为86.6%和83.7%,在嵌入式平台的平均速度为14.56 fps。更多还原

【Abstract】 Aiming at application of machine vision technology in embedded platform,a kernel correlation filter tracking model with occlusion detection and adaptive template updating is proposed to improve the robustness of scale change and occlusion when the tracker is running on embedded platforms. In order to cope with the change of target scale during the target movement,multi-scale estimation and dynamic template update strategy are added to the kernel correlation filter,At the same time,in order to ensure that the target can be detected again after occlusion,a particle filter is used to resample multiple particle candidates to detect the target. Experimental results show that the mean distance precision( mean DP) and mean overlap rate precision( mean OP) of the tracker for the OTB dataset are 86. 6 % and 83. 7 %,respectively,and the average speed on the embedded platform is14. 56 fps.更多还原