计算机应用 ›› 2015, Vol. 35 ›› Issue (6): 1795-1800.DOI: 10.11772/j.issn.1001-9081.2015.06.1795

• 行业与领域应用 • 上一篇    下一篇

基于Kinect的指尖检测与手势识别方法

谈家谱, 徐文胜   

  1. 北京交通大学 机械与电子控制工程学院, 北京 100044
  • 收稿日期:2014-12-25 修回日期:2015-04-01 发布日期:2015-06-12
  • 通讯作者: 徐文胜(1970-),男,湖北鄂州人,副教授,博士,主要研究方向:云制造、人机交互。wshxu@bjtu.edu.cn
  • 作者简介:谈家谱(1992-),男,江西南昌人,硕士研究生,主要研究方向:模式识别、图像处理.
  • 基金资助:

    国家自然科学基金资助项目(51175033)。

Fingertip detection and gesture recognition method based on Kinect

TAN Jiapu, XU Wensheng   

  1. College of Mechanical, Electronic and Control Engineering, Beijing Jiaotong University, Beijing 100044, China
  • Received:2014-12-25 Revised:2015-04-01 Published:2015-06-12

摘要:

针对基于视频的弯曲指尖点识别难、识别率不高的问题,提出一种基于深度信息、骨骼信息和彩色信息的手势识别方法。该方法首先利用Kinect相机的深度信息和骨骼信息初步快速判定手势在彩色图像中所在的区域,在该区域运用YCrCb肤色模型分割出手势区域;然后计算手势轮廓点到掌心点的距离并生成距离曲线,设定曲线波峰与波谷的比值参数来判定指尖点;最后结合弯曲指尖点特征和最大内轮廓面积特征识别出常用的12个手势。实验结果验证阶段邀请了6位实验者在相对稳定的光照环境条件下来验证提出的方法,每个手势被实验120次,12种手势的平均识别率达到了97.92%。实验结果表明,该方法能快速定位手势并准确地识别出常用的12种手势,且识别率较高。

关键词: Kinect, 肤色模型, 开源计算机视觉库, 指尖检测, 手势识别

Abstract:

The recognition of bending finger points based on video is difficult and recognition rate is not high. Focusing on the issue, a hand gesture recognition method based on depth image, skeleton image and color image information was proposed. Firstly, the gesture in color image area was initially determined rapidly by using depth information and skeleton information of Kinect, and the hand posture region was extracted from the gesture area by using the YCrCb color model. Then the distances between the gesture contour points and the palm point were calculated to generate the distance curve, and the ratio of curve peaks to troughs was set up to obtain finger point. Finally, commonly used 12 gestures were identified by combining bending finger point features and the maximum amount of contour area. Six experimenters were invited to validate the proposed method in experimental results verification phase. Every gesture was experimented 120 times under the condition of relatively stable light environment and the recognition rate of 12 gestures was 97.92% on average. The experimental results show that the proposed method can quickly location gestures and accurately recognize the commonly used 12 kinds of hand gestures, and the recognition rate is high.

Key words: Kinect, skin color model, Open Source Computer Vision Library (OPENCV), finger detection, gesture recognition

中图分类号: