一种基于Kinect与SVM的人体动作识别算法  

Research on the Action Recognition Based on Kinect and SVM Algorithm

作　　者：刘芬 吴志攀[2] LIU Fen;WU Zhi-pan(Department of Information Technology,City College of Huizhou,Huizhou 516025;School of Information Science and Technology,Huizhou University,Huizhou 516007)

机构地区：[1]惠州城市职业学院信息学院,惠州516025 [2]惠州学院信息科学技术学院,惠州516007

出　　处：《现代计算机》2019年第18期55-58,共4页Modern Computer

基　　金：惠州学院本科教学质量与改革工程项目(No.2017-15)

摘　　要：目前,传统的人机交互已经不能满足人们的需求,一种倾向于更加自然化、便捷化的人机交互方式――人体动作识别,具有重要的研究意义。提出一种基于Kinect与SVM的人体动作识别算法,先采用Kinect传感器对人体动作进行捕捉生成深度图像,经处理建立三维人体模型,提取出其运动向量的夹角和模比值作为特征值,最后使用SVM对该特征值进行人体动作分类识别。采用UCF101数据集和MATLAB程序对人体动作识别的整个过程进行验证,该算法具有很高的鲁棒性和识别率。Nowadays, the traditional human-computer interaction has been unable to meet people's needs. A more natural and convenient ones, action recognition, has the important research significance. Proposes an action recognition algorithm based on Kinect and SVM algorithm. Firstly, the Kinect sensor is used to capture the action and generate the depth image. After processing, a three-dimensional human model is established, and the ratio of the angle and the modulus of its action vector are extracted as eigenvalues. Finally, SVM is used to classify and recognize the action. The whole process of the action recognition is verified by using UCF101 dataset and MATLAB program, and the algorithm has high robustness and recognition rate.

关 键 词：人体动作识别 支持向量机 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]