摘要
指尖检测是人机交互过程中的关键技术,因为人手的差异,对指尖的检测总会存在一些误判点。文中在基于曲率算法的基础上,提出了利用凸包算法和平行向量进行指尖检测的方法。该方法首先利用Kinect获取人体的骨骼信息和深度信息图像,通过人手的关节点锁定手部位置,并利用人手肤色特征和边缘检测算法提取手部区域轮廓。然后在手部区域的轮廓上根据曲率来检测类指尖点,结合凸包计算排除凹点和手臂点,最后根据手指的两侧接近平行的特性排除弯曲的手指或者非手指,最终检测出有效的指尖。实验结果表明,该方法在复杂背景下能够对不同的类指尖点进行排除,并且有较高的检测精度。
Fingertip detection is a crucial technology in the process of human- computer interaction. Because of differences in human hands,there will always be some misjudgment points in fingertip detection. Based on curvature algorithm,a method of fingertip detection is proposed using convex-concave algorithm and parallel vector. Firstly,it obtains information of human bone and in-depth image using Kinect,locking hand position by joints of human hands,and extracts the hand contour area using color characteristics of human hand and edge detection algorithm. Then on the contour of the hand region,fingertips are detected according to the curvature,and combined with the convex hull,pits and arm points are computed and excluded. At last,according to the characteristic that both sides of the finger are nearly parallel,curved fingers or non-fingers are excluded,and valid fingertips are detected. Experimental results showthat this method can exclude different classes of fingertips under complex background with higher detection accuracy.
出处
《计算机技术与发展》
2016年第7期14-18,共5页
Computer Technology and Development
基金
上海市科学技术计划资助项目(14590500500)
关键词
指尖检测
曲率
凸包
平行向量
KINECT
fingertip detection
curvature
convex-concave
parallel vector
Kinect