基于Kinect深度信息的指尖检测

Fingertip Detection Based on Kinect Depth Image

指尖检测是目前人机交互研究的热点之一,针对基于普通摄像机的指尖检测容易受到复杂背景的影响而无法准确定位的问题,提出了基于Kinect深度信息的指尖获取方法。利用Kinect获取的深度信息将手单独分离出来,从而更好的实现手或手指与虚拟物体的交互。研究了一种利用NITE库函数定位手的位置的方法,根据获取的手的位置能够迅速而准确的从复杂背景中提取手部信息。再利用道格拉斯-普克算法得到手的轮廓曲线,最后利用轮廓分析法从凸包点中识别出指尖。实验表明该方法能够精确的定位到手的各个指尖位置,识别率达到80%,该方法实现简单,实时性好,有较好的鲁棒性。

Fingertip detection is one of the hot topics in human-computer interaction research. Aiming at the problem that fingertip detection based on ordinary camera is susceptible to the complex background. A fingertip acquisition method based on Kinect depth image is proposed,which separates the hand from the depth information obtained by Kinect,so as to better realize the interaction of hand or finger with virtual object. This paper studies a method to locate the hand position by using NITE library function. The hand information can be extracted from complex background quickly and accurately according to the position of the hand.Using Douglas-Peucker algorithm to obtain the contour curve of the hand,and finally identify the fingertip from the convex hull points using contour analysis. Experiments show that the proposed method can accurately locate the fingertips of each hand,and the recognition rate is 96.75%. The method is simple and has good robustness,also it is real-time.