基于Kinect深度图像的指尖检测与手势识别

针对基于普通摄像头的手势识别系统在不同光照条件和复杂环境下易受影响的问题，提出一种基于kinect深度图像进行指尖检测和手势识别的算法. 首先利用Kinect传感器获取深度图像，再利用OpenNI手部跟踪器检测出手部的位置，根据手部位置对手势进行深度阈值分割. 提出一种结合凸包和曲率检测指尖的算法，检测出指尖数目和位置后，计算出包括指尖和手掌水平方向的夹角、相邻两个指尖夹角以及指尖与掌心的距离的特征向量，最后利用支持向量机（SVM）对预定的9种数字手势进行识别. 实验邀请5位实验者在复杂环境下每个手势做30次，每次的手势角度不同，实验结果表明该方法能够准确检测出指尖的数目和位置，9种数字手势平均识别率达到97.1%，该方法使用特征简单，实时性好，有较好的鲁棒性.

Fingertip Detection and Hand Gesture Recognition Based on Kinect Depth Image

Aiming at the problem that hand gesture recognition system based on ordinary camera is susceptible to the different lighting conditions and complex background, a fingertip detection and hand gesture recognition algorithm based on Kinect depth image is proposed. First, we get depth image by Kinect sensor. Then the hand region is extracted by putting the depth of thresholds on hand point detected by using OpenNI library. Fingertip detection based on convex hull and curvature is proposed. After the number of fingertips and the location of fingertips being detected, it calculates a feature vector including the number of fingers, the angles between fingertips and horizontal of the hand, the angles between two consecutive fingers, and the distance between fingertips and hand center point. Finally, a support vector machine(SVM) is applied to identify nine scheduled number hand gesture. Five experimenters are invited to perform 9 different hand gestures in the complex environment, which each gesture is repeated at thirty times and the angle of hand gesture is different every time. The experiment results show that this algorithm can detect the number and location of fingertips, and the recognition rate of nine hand gesture is 97.1% on average. This proposed method uses simple features and has good robustness, also it is real-time.