基于Kinect深度图像的手势识别分类  

Gesture Recognition Classification Based on Kinect Depth Image

作　　者：潘峥嵘[1] 杜新怡 PAN Zheng-rong;DU Xin-yi(College of Electrical and information Engineering,Lanzhou University of Technology,Lanzhou 730050 China)

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050

出　　处：《自动化技术与应用》2019年第4期143-147,共5页Techniques of Automation and Applications

摘　　要：针对肤色识别容易受到环境、光照、类肤色等复杂背景干扰的问题,提出了一种基于深度图像的手势识别方法。该方法首先通过Kinect深度摄像头获取深度图像,再利用开源库OpenNI获得手部的粗略位置,根据手部位置的距离值对手势进行深度阈值分割,同时跟踪手部。利用OpenCV相关函数对分割后的图像进行形态学预处理,获得手的外部轮廓坐标,根据手的重心及指尖到重心的距离得到指尖坐标并对三个常用交互手势"石头"、"剪刀"、"布"进行分类。该方法所使用的特征简单,识别的实时性好,有较好的鲁棒性,为人机交互提供了一种方法。Aiming at the problem that skin color recognition is easily disturbed by complex background such as environment, light and skin color, a method of gesture recognition based on depth image is proposed. The method firstly obtains the depth image through the Kinect depth camera, and then uses the open source library OpenNI to obtain the rough position of the hand, and performs depth threshold segmentation of the gesture according to the distance value of the hand position, and simultaneously tracks the hand. Morphological processing of the segmented image using OpenCV correlation function. Then, it obtains the external outline coordinates of the hand. The coordinates of the fingertips are obtained according to the center of gravity of the hand and the distance from the fingertip to the center of gravity, and three commonly used interactive gestures "stone","scissors","Cloth" are classified. classification. The method has the simple characteristics, good real-time recognition, and good robustness, provides a method for human-computer interaction.

关 键 词：深度图像 指尖检测 手势识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]