Abstract:

Correlation filter has been widely used in object tracking because of its superiority with efficiency and robustness, but it does not deal with occlusion and scale variation well. A fusion correlation particle filter object tracking algorithm is proposed, more target information and background information can be learned by the multiple correlation filters, it helps to distinguish the target and background better. And the particle filter sampling strategy is presented to capture target quickly when the target leaves the occlusion area. For the scale estimation, multi-scale coefficients are used to determine the scale variation of target, the factor corresponding to the largest response value of candidate region and filter is selected as the scale coefficient. In addition, as the number of particles increases, the computation of particle filtering algorithm also increases, to ease this problem, a re-sampling algorithm based on particle proliferation is proposed, which can make for the tracking efficiency. Finally, three experiments are performed to verify the effectiveness and robustness of the proposed algorithm in dealing with occlusion and scale variation.

Key words: object tracking, correlation filter, particle filter, scale estimation, occlusion

摘要：

相关滤波算法因其优越的高效性和鲁棒性被广泛应用于目标跟踪领域，但是该算法无法很好地处理目标遮挡和尺度变化等问题。针对该现象，提出了一种融合相关粒子滤波目标跟踪算法，该算法采用多个相关滤波器，学习到更多目标信息和背景信息，提高了目标与背景辨识度，并且引进了粒子滤波随机采样策略，在目标离开遮挡物时能够快速捕捉到目标。在尺度估计中引入了多尺度因子，对定位到的目标进行多尺度缩放，选用与滤波器响应值最大区域对应的尺度因子作为缩放比例，从而对目标进行尺度更新；粒子滤波算法随着粒子数目的增加，其计算量也随着增加，针对该问题，提出了基于粒子繁衍的重采样算法，在跟踪效率上做了提升。对提出的算法进行了三部分对比实验，实验结果验证了提出算法在处理目标遮挡和尺度变化问题上的有效性。

关键词: 目标跟踪, 相关滤波, 粒子滤波, 尺度估计, 目标遮挡