融合深度与肤色特征的自适应手部跟踪算法

手部跟踪技术是实现自然人机交互的关键。针对现有跟踪方法易受光照、环境等影响及鲁棒性差的不足,提出一种融合深度与肤色特征的自适应手部跟踪算法。考虑手部运动过程的形变,该算法首先利用深度平滑连续性选取深度阈值以实现跟踪区域的自适应尺度变化,获得手部候选区域。在此基础上建立YCbCr空间肤色归一化直方图,在粒子滤波框架下将跟踪问题转换为贝叶斯估计问题,基于最大后验准则确定手部位置,并通过监测粒子重要性权值的方差解决跟踪失效问题,实现复杂观测环境下的鲁棒跟踪。实验结果表明,该跟踪算法可适应不同复杂环境,鲁棒性良好。

Adaptive Hand Tracking Based on Fusing Depth and Skin Features

Hand tracking is key to realize natural human-computer interaction.The existing hand tracking algorithms for human-computer interaction can be easily affected by illumination and have poor robustness.In this paper,an adaptive hand tracking by the fusion of depth and skin color features was put forward.Considering the deformation of hand,the algorithm uses the depth threshold to achieve adaptive changes of tracking region with the continuity and smoothness of depth feature and obtains the hand alternative regions.Then,via skin color feature of YCbCr space,the skin normalized histograms were established to describe the alternative regions.Under the framework of particle filter,the hand tracking is converted into Bayes estimation problem and the algorithm determines the hand position based on the maximum a posterior (MAP).The variance of particle weights was monitored to solve the problem of tracking failure.Experimental results show that,the proposed algorithm can achieve accurate and robust tracking performance in real-time with complex background.