基于Kinect的特定人员鲁棒识别与定位

Robust recognition and localization of specific persons using kinect

为避免人机共处时相撞,提出了一种基于Kinect视觉的特定人员识别与定位方法.利用Kinect传感器深度图像中的用户索引位信息、同步的彩色图像与深度图像的对应关系,滤除彩色图像中的复杂背景,确定感兴趣目标区域.进而在彩色图像内,基于加速鲁棒特征算法(SURF)在连续帧间完成特征匹配,快速识别特定目标人员;在深度图像内,采用无迹卡尔曼滤波(UKF)实现了在摄像机坐标系内的人员定位.实验结果验证了算法的鲁棒性,表明算法能够适应光线变化、目标再入和部分遮挡等情况.

A Kinect-based person recognition and localization method is proposed to avoid collisions during the co-extstence of human and robots. Firstly, by using the player index data in Kinect depth images and the correspondence between synchronized depth and RGB images, the background in RGB image is rejected to attain the interest target area. Secondly, Speeded-Up Robust Features （SURF） algorithm is used to detect features rapidly in sequential RGB images and then recognize the specific person. Finally, the person＇s position is estimated using Unscented Kalman Filter （UKF）. Experiential results verify the algorithm robustness and show that the method can adapt to those cases of light changing, target re-entering and partial occlusion.