Abstract: The proposed method embeds mean-shift into the tracking frame of the particle filter algorithm.The authors regard color distribution as the observation model.The components of HSV color space are divided into unequal intervals according to human color perception.Then we use the color histograms based on kernel function to build the model.The algorithm overcomes the default of both particle filter and mean-shift.The former has large computation cost.The latter can easily trapped into local maximum and can not recover from the error.Experimental results show that the proposed algorithm has real-time property and robustness.

Key words: object tracking, particle filter, mean-shift, kernel function, HSV

摘要： 将均值漂移算法嵌入到粒子滤波的跟踪框架中，将颜色分布作为观测模型，将HSV颜色空间根据人类的颜色感知差异，对各个分量进行非等间隔量化，然后利用基于核函数的直方图进行建模。该算法克服了粒子滤波计算量较大的缺点，同时也克服了均值漂移算法容易陷入局部最大且无法恢复的缺点。实验结果表明，该方法具有较强的实时性和鲁棒性。

关键词: 目标跟踪, 粒子滤波, 均值漂移, 核函数, HSV