基于复合特征和动态阈值圆法的手势识别算法研究

针对传统手势识别中用肤色分割手部区域效果的局限性，采用Kinect获取深度信息来分割手掌，能得到较好的效果。对手掌轮廓进行多边形逼近，将凸包点作为候选指尖点。利用非零像素（白）到最近零像素的距离提取掌心，用线性回归动态调整阈值圆半径，将无用凸包点过滤，实现指尖点的准确提取。在分类识别中，将图像的Hu矩和指尖点个数组合起来，作为复合手势特征，导入KNN分类器中，实现手势识别。实验证明，基于复合特征和动态阈值圆法的手势识别算法具有较好的识别率和实时性。

Gesture recognition algorithm based on combined features and dynamic threshold circle

For the limitation of hand region segmention with skin color in the traditional gesture recognition, this paper used Kinect to obtaining depth information to segment the palm could get better results. The palm contour was polygonal approximation, and selscted the convex points as the candidate fingertip. It distinguished the fingertips accurately from the unwanted convex points by a threshold circle. Utilizing the distance relation of the non zero pixel(white) to the nearest zero pixel to extract the palm center, the threshold circle transformed radius dynamically by linear regression method. In the classification and recognition, combining the Hu moment and the number of fingertip as the feature of gesture, and the KNN classifier achieved gesture recognition. Experiments show that the gesture recognition algorithm based on composite features and dynamic threshold circle method has better recognition rate and real-time performance.