查询结果:   朱正伟,祝磊,饶鹏.基于BS-HMM和巴式距离的手势识别技术研究[J].计算机应用与软件,2019,36(6):163 - 166,253.
中文标题
基于BS-HMM和巴式距离的手势识别技术研究
发表栏目
人工智能与识别
摘要点击数
875
英文标题
GESTURE RECOGNITION BASED ON BS-HMM AND BHATTACHARYYA DISTANCE
作 者
朱正伟 祝磊 饶鹏 Zhu Zhengwei Zhu Lei Rao Peng
作者单位
常州大学信息科学与工程学院 江苏 常州 213164 常州光电技术研究所 江苏 常州 213164    
英文单位
School of Information Science and Engineering,Changzhou University,Changzhou 213164, Jiangsu,China Changzhou Institute of Optoelectronic Technology, Changzhou 213164, Jiangsu, China    
关键词
手势识别 贝叶斯感知隐马尔科夫模型 巴氏距离 HON4D特征 HOG特征
Keywords
Gesture recognition BS-HMM Bhattacharyya distance HON4D feature HOG feature
基金项目
国家自然科学基金项目(61772090)
作者资料
朱正伟,教授,主研领域:智能检测技术及应用。祝磊,[HTSS]硕士生。饶鹏,[HTSS]研究员。 。
文章摘要
开发一个基于深度图像的手势识别系统,将巴氏距离(Bhattacharyya distance)引入到贝叶斯感知隐马尔科夫模型(BS-HMM)中,称为BDBS-HMM。使用深度摄像机Kinect捕获深度序列图,通过骨架信息对手部位置进行跟踪,识别手部区域,得到手部分割图;从分割图像中提取4D曲面法线方向分布(HON4D)特征和方向梯度直方图(HOG)特征表示运动模式;将每k个连续的特征向量组合成一个序列分布变换所有训练特征向量,使用分布序列来对BDBS-HMM进行训练。该系统在使用MSRGesture3D数据库和自己建立的数据库的情况下,将BDBS-HMM与标准HMM和BS-HMM进行比较,实验结果表明了该系统的优越性。
Abstract
In this paper, we developed a gesture recognition system based on depth image, and Bhattacharyya distance was incorporated into BS-HMM,named BDBS-HMM. The depth sequence image was captured by the depth camera Kinect, and the hand segment image was obtained by tracking and recognizing the hand region through the skeleton information. Then the HON4D features and HOG features were extracted from the segmentation image to represent the motion pattern. Each k consecutive feature vectors were combined into a sequence distribution to transform all training feature vectors. Distribution sequence was used to train BDBS-HMM. The system was compared with the standard HMM and BS-HMM in the case of using MSRGesture3D database and our database. The experimental results show the superiority of the system.
下载PDF全文