首页 | 本学科首页   官方微博 | 高级检索  
     

基于深度数据的人体动作识别方法
引用本文:王鑫,沃波海,管秋,陈胜勇.基于深度数据的人体动作识别方法[J].中国图象图形学报,2014,19(6).
作者姓名:王鑫  沃波海  管秋  陈胜勇
作者单位:浙江工业大学计算机科学与技术学院,浙江工业大学计算机科学与技术学院,浙江工业大学计算机科学与技术学院,浙江工业大学计算机科学与技术学院
基金项目:国家自然科学基金(61303142,61173096,61103140),教育部高等学校博士学科点专项科研基金(20113317110001),浙江省自然科学基金项目(Y1110882,Y1110688,R1110679,LY13F020034),浙江省教育厅一般科研项目(Y201330304)
摘    要:本文提出了一个基于流形学习的动作识别框架,用来识别深度图像序列中的人体行为。本文从Kinect设备获得的深度信息中评估出人体的关节点信息,并用相对关节点位置差作为人体特征表达。在训练阶段,本文利用Lapacian eigenmaps(LE)流形学习对高维空间下的训练集进行降维,得到低维隐空间下的运动模型。在识别阶段,本文用最近邻差值方法将测试序列映射到低维流形空间中去,然后进行匹配计算。在匹配过程中,通过使用改进的Hausdorff距离对低维空间下测试序列和训练运动集的吻合度和相似度进行度量。本文用Kinect设备捕获的数据进行了实验,取得了良好的效果;同时本文也在MSR Action3D数据库上进行了测试,结果表明在训练样本较多情况下,本文识别效果优于以往方法。实验结果表明本文所提的方法适用于基于深度图像序列的人体动作识别。

关 键 词:Kinect  sensor  人体动作识别  流形学习  Hausdorff距离  深度数据

Human action recognition based on depth image data
wangxin,wobohai,guanqiu and chenshengyong.Human action recognition based on depth image data[J].Journal of Image and Graphics,2014,19(6).
Authors:wangxin  wobohai  guanqiu and chenshengyong
Affiliation:College of Computer Science and Technology, Zhejiang University of Technology,College of Computer Science and Technology, Zhejiang University of Technology,College of Computer Science and Technology, Zhejiang University of Technology
Abstract:Human action recognition is a widely studied area in computer vision and machine learning, it has many potential applications including human computer interfaces, video surveillance, health care. In the past decade, extensive research efforts focused on recognizing human action from monocular video sequences. Since human motion is articulated, and capturing human joint characters accurately from video is a very difficult task. The recent introduction of real time depth cameras such as Kinect sensor, give us the choice to use 3D depth data of the scene instead of the picture. In this paper, we present a manifold-based framework for human action recognition using depth image data captured from depth camera. With the recent release of Kinect sensor and the technology assessing skeleton joint position from depth image matured, recent research used 3D skeleton joint position information as human body representation and achieved good recognition performance. As we know, human action is composed of ordered posture set, and the difference between postures is only few changes of 3D joints pairwise, most of 3D that changes little. This paper estimated the 3D joint locations from Kinect depth images and used pairwise relative positions as the representation of human features. In training phase, This paper use Lapacian Eigenmaps(LE) to build action model in low dimension space. In test phase, nearest-neighbor interpolation technique is used to map test sequence to manifold space, then measure distance with test sequence and train data. A novel modified Hausdorff distance is employed to measure similarity and fitness of test sequence and train data in matching process. The recognition performance of proposed method was evaluated from Kinect sensor dataset and the result confirmed the proposed method can work well in several experiments. This paper also tested proposed method on the MSR Action3D dataset and achieved state of the art accuracy in our comparison with related work when train set has many samples.Manifold learning is a effective nonlinear dimensionality reduction method and low-dimensional motion models can be trained well when train sample size is large. This paper proposes a novel human action recognition based on manifold learning. The experimental results show the effectiveness of the proposed method for human action recognition based on depth image sequence.
Keywords:Kinect sensor  human action recognition  manifold learning  Hausdorff distance  Depth data
点击此处可从《中国图象图形学报》浏览原始摘要信息
点击此处可从《中国图象图形学报》下载免费的PDF全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号