基于双层分类模型的人体动作识别方法  

Recognition method of human activity using double-layer classification model

作　　者：赵雪章[1] 席运江[2] 黄雄波[1] ZHAO Xue-zhang;XI Yun-jiang;HUANG Xiong-bo(Department of Electronic Information, Foshan Polytechnic, Foshan 528137, China;School of Business Administration, South China University of Technology, Guangzhou 510000, China)

机构地区：[1]佛山职业技术学院电子信息系,广东佛山528137 [2]华南理工大学工商管理学院,广东广州510000

出　　处：《计算机工程与设计》2018年第12期3860-3866,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(71371077);佛山市科技计划基金项目(2015AB004241)

摘　　要：针对现有人体动作识别方法不能很好地识别静态图像中动作的问题,提出一种结合姿态库和AdaBoost双层分类模型的识别方法。利用人体位置数据和Hausdorff距离创建初始姿态库,利用第一层分类器进行训练,获得由特征向量和空间信息组成的姿态库;基于第一层分类器获得测试图像的空间姿态激活向量(SPAV);将SPAV进行级联作为第二层分类器的输入,以具有最大分类分数的动作类型作为判别结果。在PASCAL2010和KTH数据集进行相关实验,实验结果表明,该方法在识别精确性和处理时间上具有优越的性能。For the issues that the existing human action recognition method can’t recognize the action from the static image well,an identification method combining posture set and AdaBoost double-layer classification model was proposed.The initial position database was created using the human position data and the Hausdorff distance,and trained by the first classifier to obtain the posture set composed of the eigenvector and the spatial information.The spatial posture activation vector(SPAV)of the image was obtained based on the first layer classifier.The SPAV was cascaded and used as input of the second classifier,and the action type with the largest classification score was taken as the result of the discrimination.The results show that the proposed method has superior performance in recognition accuracy and processing time.

关 键 词：人体动作识别 AdaBoost双层分类模型 HAUSDORFF距离 姿态库 空间姿态激活向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]