基于超像素与BoF的运动目标跟踪算法

摘要

针对由遮挡、光照、形变等干扰产生的漂移问题,提出基于超像素和BoF的运动目标跟踪算法.此算法首先利用SLIC方法对运动目标的观测区域进行超像素分割,然后基于超像素构造中层视觉线索的超像素字典以及低层像素特征的BoF字典,实现对运动目标表观模型的混合建模,最后引入粒子滤波框架和在线字典更新,以适应目标和背景的变化.实验结果表明,该算法能够很好地应对严重遮挡、非刚性变换、复杂背景等干扰因素的影响,具有很好的稳定性和鲁棒性.%In allusion to the drift problems caused by occlusion,illumination and deformation,a motion target tracking al-gorithm based on superpixel and BoF is proposed. The SLIC method is adopted to perform superpixel segmentation for observa-tion area of motion target. The superpixel dictionary with mid-level visual cues and BoF dictionary with low-level superpixel characteristic are constructed based on superpixel to realize hybrid modeling for motion target appearance model. Particle filter-ing framework and online dictionary update are introduced to adapt the variations of targets and background. The experimental re-sults show that the algorithm can tackle the effects of severe occlusion,non-rigid deformation,complex background and other in-terference factors,and has good stability and robustness.