基于Camshift和Kalman滤波的仿人机器人手势跟踪

对仿人机器人MIR-1的双目视觉系统实现实时手势跟踪.通过颜色直方图反投影,将每帧RGB输入图像转换为二维的肤色概率分布图像,基于Camshift算法计算手势跟踪窗口的位置和大小,并用Kalman滤波预测手心位置,有效地解决了背景中大面积肤色干扰和手势部分被遮挡等问题.在仿人机器人MIR-1上完成的手势跟踪实验,验证了此方法的实用性和有效性.

The Hand Tracking for Humanoid Robot Using Camshift Algorithm and Kalman Filter

Based on Camshift(Continuously Adaptive Mean Shift) algorithm and the constant-velocity Kalman filter algorithm,a real-time hand tracking system for the stereo vision system of humanoid robot MIR-1 was presented.For each video frame,the raw image is converted to a 2-D skin color probability distribution image via the skin color histogram back-projection.Camshift algorithm is then applied to find the mode(peak) of the probability distribution,which estimates the center and size of tracking window.To deal with the hand occlusion by other objects and choose the correct skin region when multiple image regions are skin colored,a two-dimensional Kalman filter(stochastic estimator) is used to track the hand region centroid.The experimental results show that this algorithm is robust and efficient.