基于深度信息与DBN的手势检测与识别

Gesture Detection and Recognition Based on Depth Information and DBN

提出基于深度信息与深度置信网络(DBN)的手势检测与识别算法.首先通过Kinect提供的深度信息定位手势区域,并找出手势区域的方向线,根据方向线的角度把手势旋转成垂直向上.结合手掌形状特征实现手势分割得到二值化的手势图像.在此基础上,利用深度置信网络进行识别.通过深度信息,去除了复杂背景的干扰.利用方向线使得手势具有抗旋转性.利用深度置信网络无需其他特征提取.实验结果表明,本文检测手势区域准确,计算量较小,能够快速地实现手势识别,并能够获取较高的识别率.

A gesture detection and recognition algorithm based on the depth information and deep belief network is presented. First, the hand area is located by the depth information obtained by Kinect. Then the hand gesture is rotated by the angle of direction line and segmented by the features of hand＇s shape. Last the gestures are recognized by DBN. In this method, the obstructions of background are removed. The gestures have certain ability of geometric distortions by direction line. The use of DBN makes no other feature extraction. Experimental results show that this method can detect hand area correctly and is easy to get a high recognition rate with low computational load.