基于手掌分割的摄像机阵列手部定位技术研究

Research on camera array hand position detection technique based on palm segmentation

人手的三维空间定位是虚拟现实交互中的关键技术之一。人手目标是具有多自由度的柔性体,而且运动范围较大,难以准确定位。提出一种基于手掌分割的摄像机阵列手部定位技术,可以有效解决上述两大难点。该方法首先对单摄像机采集到的图像进行基于颜色的手部整体轮廓信息提取。然后,通过分析轮廓特征分割手掌。以手掌为近似刚体,在二维图像中对手掌进行定位。最后,智能选取摄像机阵列中符合定位要求的多个摄像机,应用摄像机配对、投影几何以及最小二乘法估计人手目标的三维位置。实验结果表明,该方法抗干扰性好,对不同手型的适应能力强。同时应用多摄像机进行手部定位,不仅跟踪范围广,而且可以在标定困难的环境中,应用未标定的摄像机阵列获得较高的精度。

3D hand position detection is a crucial technology in virtual reality interactive system. Human hand is a multiple degree of freedom soft body and has a wide range of motion. So human hand locating is difficult. This paper proposes a new method, which locates human hand with a camera array based on palm segmentation to solve these problems. Firstly, we extract the contour of hand based on the color information of the image acquired with single camera. Secondly, we segment the palm image through analyzing the hand contour, and take the palm as a rigid body and locate the palm in 2D image. Finally, we use local spatial outlier detecting algorithm to select the valid cameras in camera array and apply camera pairing, projective geometry and least square method to calculate the 3D position of the hand. Experiment result shows that this method has a good anti-interference characteristic and strong adaptive capacity for different hand gesture. Using camera array, this method can detect hand position in wide tracking region and get precise result with uncalibrated camera array in an environment with calibration difficulty.