基于Kinect骨骼信息与深度图像的指尖点检测
|
摘要
针对普通摄像头手势识别系统易受复杂环境和光照条件等因素影响,存在对指尖点的漏判、误判问题,提出一种基于Kinect骨骼信息与深度图像的掌心点提取和指尖点检测的手势识别方法;在DRVI平台上创建Kinect的接口控件,对Kinect传感器获取人体骨骼信息和深度图像进行分析,采用了坐标映射、图像分割、距离变换的关键技术和方法从深度图中分割出手势部分区域,对手势区域形态学处理,结合凸包和K-曲率算法检测不同手势中指尖点的个数和位置,计算不同手势凸包轮廓上的点集生成的HOG (Histogram of Oriented Gradient)特征描述子,最后利用特征描述子对预定的6种数字手势进行识别;经实验测试可以在复杂环境和不同光照情况下正确识别指尖点。
The hand gesture recognition system is susceptible to the light conditions and complex environments by using ordinary camera,so the fingertips are always missed and misjudged.To solve these problems,a new method for fingertip detection and palm point extraction of hand gesture is proposed based on Kinect skeleton information and depth image.Creating a Kinect control on the DRVI platform and analyzing the information of the skeleton and depth image is acquired by Kinect sensor,the technology and method of coordinate mapping,image segmentation and distance conversion can be used to segment the hand area on the depth image,which need morphological processing.After the number of fingertips and the location for hand gesture being detection with combining convex hull and K-curvature algorithm,calculating the HOG feature descriptor generated by the point set on the contours of different gesture hulls.Finally,the HOG feature descriptor is applied to identify six scheduled number hand gesture.The experiment results show that the proposed method can identify fingertips in complex environments and different lighting conditions.
The hand gesture recognition system is susceptible to the light conditions and complex environments by using ordinary camera,so the fingertips are always missed and misjudged.To solve these problems,a new method for fingertip detection and palm point extraction of hand gesture is proposed based on Kinect skeleton information and depth image.Creating a Kinect control on the DRVI platform and analyzing the information of the skeleton and depth image is acquired by Kinect sensor,the technology and method of coordinate mapping,image segmentation and distance conversion can be used to segment the hand area on the depth image,which need morphological processing.After the number of fingertips and the location for hand gesture being detection with combining convex hull and K-curvature algorithm,calculating the HOG feature descriptor generated by the point set on the contours of different gesture hulls.Finally,the HOG feature descriptor is applied to identify six scheduled number hand gesture.The experiment results show that the proposed method can identify fingertips in complex environments and different lighting conditions.
引文
[1]高晨,张亚军.基于Kinect深度图像的指尖检测与手势识别[J].计算机系统应用,2017.23(6):193-197.
[2]杜京义,胡益民,刘宇程.基于区域分块的SIFT图像匹配技术研究与实现[J].光电工程,2013,40(8):52-58.
[3]刘小建,张元.基于Kinect的手势识别及其在场景驱动中的应用[D].太原:中北大学,2017.
[4]Simen Andresen,Martin Stokkeland,Vegar qsthus.Hand Detection Using Color Recognition[J].Object Tracking and Gesture Recognition-Shortened Version.2013,29(6).
[5]方华,刘诗雄,田敬北.基于Kinect骨骼系统的手势识别研究[J].2015,22(5):65-67.
[6]罗元,杨明珠,张毅.基于改进去伪匹配SURF算法的静态手语字母识别[J].激志,2014,35(7):1-4.
[7]Cui Y T,Weng J J.Viewased hand segmentation and hand-sequence recognition with complex backgrounds[A].In Proceedings of the IEEE International Conference on Pattern Recognition[C].Osaka,Japan,1997:617-621.
[8]宋海生,刘平和,王全州,等.基于人体骨骼和深度图像信息的指尖检测方法[J].计算机工程与科学,2014,9.
[9]黄山,罗琳.基于视觉的手势识别系统关键技术研究[D].南京:东南大学,2015.
[10]张毅,张烁,罗元,等.基于Kinect深度图像信息的手势轨迹识别及应用[J].计算机应用研究,2012,9.
[11]李长龙,董秀成.基于Kinect深度图像的手势识别研究[D].成都:西华大学,2014.
[12]毛雁明,张立亮.基于Kinect深度信息的手势分割与识别[J].系统仿真学报,2015,4.
[13]Zhang Jinyu,Yan Chen,Huang Xianxiang.Edge detection of images based on improved sobel operator and genetic algorithms[A].Proceedings of2009International Conference on Image Analysis and Signal Processing[C].2009:32-35.
[14]Shotton,J.Fitzgibbon A.Cook M.et al.Real-time human pose recognition in parts from single depth images[A].2011IEEE Conference on Computer Vision and Pattern Recognition(CVPR)[C].2011,1297-1304.
[15]周凯,马行.基于肤色和SVM的手势识别及其应用研究[D].银川:北方民族大学,2016.
[16]郑斌珏,赵疗英.基于Kinect深度信息的手势识别[D].杭州:杭州电子科技大学,2014.
[17]陈祖雪,马苗.基于深度卷积神经网络的手势识别研究[D].西安:陕西师范大学,2016.
[18]董立峰,阮军.基于Hu矩和支持向量机的静态手势识别及应用[D].武汉:武汉理工大学,2012.
[2]杜京义,胡益民,刘宇程.基于区域分块的SIFT图像匹配技术研究与实现[J].光电工程,2013,40(8):52-58.
[3]刘小建,张元.基于Kinect的手势识别及其在场景驱动中的应用[D].太原:中北大学,2017.
[4]Simen Andresen,Martin Stokkeland,Vegar qsthus.Hand Detection Using Color Recognition[J].Object Tracking and Gesture Recognition-Shortened Version.2013,29(6).
[5]方华,刘诗雄,田敬北.基于Kinect骨骼系统的手势识别研究[J].2015,22(5):65-67.
[6]罗元,杨明珠,张毅.基于改进去伪匹配SURF算法的静态手语字母识别[J].激志,2014,35(7):1-4.
[7]Cui Y T,Weng J J.Viewased hand segmentation and hand-sequence recognition with complex backgrounds[A].In Proceedings of the IEEE International Conference on Pattern Recognition[C].Osaka,Japan,1997:617-621.
[8]宋海生,刘平和,王全州,等.基于人体骨骼和深度图像信息的指尖检测方法[J].计算机工程与科学,2014,9.
[9]黄山,罗琳.基于视觉的手势识别系统关键技术研究[D].南京:东南大学,2015.
[10]张毅,张烁,罗元,等.基于Kinect深度图像信息的手势轨迹识别及应用[J].计算机应用研究,2012,9.
[11]李长龙,董秀成.基于Kinect深度图像的手势识别研究[D].成都:西华大学,2014.
[12]毛雁明,张立亮.基于Kinect深度信息的手势分割与识别[J].系统仿真学报,2015,4.
[13]Zhang Jinyu,Yan Chen,Huang Xianxiang.Edge detection of images based on improved sobel operator and genetic algorithms[A].Proceedings of2009International Conference on Image Analysis and Signal Processing[C].2009:32-35.
[14]Shotton,J.Fitzgibbon A.Cook M.et al.Real-time human pose recognition in parts from single depth images[A].2011IEEE Conference on Computer Vision and Pattern Recognition(CVPR)[C].2011,1297-1304.
[15]周凯,马行.基于肤色和SVM的手势识别及其应用研究[D].银川:北方民族大学,2016.
[16]郑斌珏,赵疗英.基于Kinect深度信息的手势识别[D].杭州:杭州电子科技大学,2014.
[17]陈祖雪,马苗.基于深度卷积神经网络的手势识别研究[D].西安:陕西师范大学,2016.
[18]董立峰,阮军.基于Hu矩和支持向量机的静态手势识别及应用[D].武汉:武汉理工大学,2012.