双层特征优化的视觉运动目标跟踪算法

Visual tracking of moving object based on double layer features optimization

视觉监控中运动目标跟踪容易受到遮挡、目标快速运动与外观变化等因素的素影响,单层特征难以有效解决这些问题。为此,提出一种像素级与区域级特征组合优化的视觉跟踪算法。首先在像素级利用目标和背景区域颜色特征的后验概率对目标与背景进行初步判别;然后对候选区域进行超像素分割,并依据像素级的判断结果,在超像素区域内利用投票决策模型对目标与背景信息进行统计分析,得到精确的目标位置分布;最后结合均值漂移迭代搜索得到目标的准确位置,并利用双层判别结果对目标跟踪过程的遮挡情况进行检测,同时动态更新目标以及背景区域信息以适应目标外观与场景变化。与典型算法进行对比的实验结果表明,本文算法能够有效应对目标遮挡与快速运动等因素的影响,适用于复杂场景条件下实时的运动目标跟踪。

As single layer feature cannot efficiently reduce the disturbances of occlusion, fast motion and target appearance change on moving object tracking in video surveillance, an algorithm combining the features of pixel and region layer is proposed in this paper. At first, the object and background are coarsely discriminated by the posterior probability of the color feature in pixel level. Then,the candidate regions are segmented by the superpixel algorithm. Furthermore, accurate distribution of the object is provided by voting in the superpixel regions with the results of the pixel layer. Finally, the location of the object is obtained by mean shift iteration. And the occlusion is found with the discriminative mask of the pixel and superpixel layer. To adapt to the change of the object appearance and scene,fused with the detection of occlusion,the histogram of the target and background parts are dynamically updated. Experi mental results show that the proposed algorithm is applicable to real time moving object tracking in lowcontrast scenes and copes with the influence of object occlusion and fast motion efficiently.