摘要
以往的时空兴趣点算法需考虑多层时空尺度计算,复杂度高,为此结合旋度的兴趣点算法和隐马尔科夫模型提出了一种新的人体动作识别算法。从光流场中获取时空兴趣点,利用光流场的旋度找出人体相邻关节的空间信息,进一步为每一类动作训练一个时空兴趣点的隐马尔科夫模型,并采用随机选择窗口尺寸避免多尺度遍历节省计算时间。将提出的算法应用到KTH和Hollywood-2运动数据集上进行实验,结果表明,相比其他几种较新的人体动作识别算法,提出的算法在计算时间和识别能力上均具有优越性。
The previous spatio-temporal interest point algorithm is necessary to consider the high computational complexity of multilayer spatio-temporal scales.In order to solve the problem,a novel algorithm of human body motion recognition is proposed based on the roll-up interest point algorithm and hidden Markov model.Firstly,the spatio-temporal interest points are obtained from the optical flow field,and then the spatial information of the adjacent joints of the human body is found by using the curl of the optical flow field,and a hidden Markov model of spatio-temporal interest points is further trained for each type of motion.Selecting the window size randomly avoids multi-scale traversal to save computing time.The proposed algorithm is applied to the KTH and Hollywood-2motion data sets.The results show that compared with other newer human motion recognition algorithms,the proposed algorithm has superiority in both computation time and recognition ability.
作者
赵雪章
居华倩
席运江
ZHAO Xuezhang;JU Huaqian;XI Yunjiang(Department of Electronic Information,Foshan Polytechnic,Foshan 528137;School of Business Aduinstration,South China University of Technology,Guangzhou 510000)
出处
《微型电脑应用》
2018年第12期1-4,10,共5页
Microcomputer Applications
基金
国家自然科学基金面上项目(71371077)
佛山市科技计划项目(2015AB004241)
关键词
人体动作识别
时空兴趣点
隐马尔科夫模型
光流场
空间信息
Human motion recognition
Space-time interest points
Hidden Markov model
Optical flow field
Spatial information
作者简介
赵雪章(1972-),男,南阳人,副教授,硕士,研究方向:图像处理、模式识别等;居华倩(1983-),女,武穴人,讲师,硕士,研究方向:图像处理、数字媒体艺术等;席运江(1973-),男,南阳人,副教授,博士,研究方向:智能算法等.
