基于运动特征与序列袋的人体动作识别
|
摘要
针对当前的动作识别算法难以有效识别复杂动作的问题,提出一种人体行为识别算法。利用改进的密集轨迹,将视频表示为多个基础动作序列(primitive actions,PA),将其编码为PA的特征序列,通过仿射传播,将其变为索引序列;基于序列模式挖掘(sequential pattern mining,SPM),形成不同的序列袋(bag of sequence,BOS)模型;对BOS模型进行学习,计算其序列比对特征、外观匹配特征、序列集特征,构成动作的评分函数;引入线性判别分析(linear discriminat analysis,LDA),对动作的评分值进行分类学习,完成动作识别。在MSR3D与UCF-Sport数据集上进行测试,实验结果表明,面对各种复杂动作,所提算法具有更高的识别精度与稳定性。
Aiming at the problem that it is difficult to recognize complex actions effectively in the current action recognition algorithm,a complex motion recognition algorithm based on motion feature coupling bag of pocket was proposed.The video was represented as multiple PA(primitive actions)sequences using an improved dense trajectory,which encoded a sequence of features.By affine propagation,the feature sequence was transformed into PA index sequence.The resulting PA index sequences were combined by sequential pattern mining(SPM)to form a different sequence of BOS.The BOS model was learnt,its sequence alignment features,appearance matching features,sequence set features were calculated.Linear discriminate analysis(LDA)was introduced to classify learning and complete the action recognition.MSR3 Dand UCF-Sport data were used to test the algorithm,experimental results show that the proposed algorithm has higher recognition accuracy and stability in the face of complex actions.
Aiming at the problem that it is difficult to recognize complex actions effectively in the current action recognition algorithm,a complex motion recognition algorithm based on motion feature coupling bag of pocket was proposed.The video was represented as multiple PA(primitive actions)sequences using an improved dense trajectory,which encoded a sequence of features.By affine propagation,the feature sequence was transformed into PA index sequence.The resulting PA index sequences were combined by sequential pattern mining(SPM)to form a different sequence of BOS.The BOS model was learnt,its sequence alignment features,appearance matching features,sequence set features were calculated.Linear discriminate analysis(LDA)was introduced to classify learning and complete the action recognition.MSR3 Dand UCF-Sport data were used to test the algorithm,experimental results show that the proposed algorithm has higher recognition accuracy and stability in the face of complex actions.
引文
[1]SHAO Yanhua,GUO Yongcai,GAO Chao. Human action recognition using multi-feature fusion[J].Journal of Opto Electronics·Laser,2014,25(9):1818-1823(in Chinese).[邵延华,郭永彩,高潮.基于特征融合的人体行为识别[J].光电子·激光,2014,25(9):1818-1823.]
[2]Feng Z,Yang B,Li Y,et al.Real time oriented behavior driven 3Dfreehand tracking for direct interaction[J].Pattern Recognition,2013,46(2):590-608.
[3]Yuan Yuan,Zheng Xiangtao,Lu Xiaoqiang.A discriminative representation for human action recognition[J].Pattern Recognition,2016,59(12):88-97.
[4]TIAN Guohui,YIN Jianqin,HAN Xu.A novel human activity recognition method using joint points information[J].ROBOT,2014,36(3):285-292(in Chinese).[田国会,尹建芹,韩旭.一种基于关节点信息的人体行为识别新方法[J].机器人,2014,36(3):285-292.]
[5]YIN Jianqin,LIU Xiaoli,TIAN Guohui.Human action recognition based on the sequence of key points[J].ROBOT,2016,38(2):200-207(in Chinese).[尹建芹,刘小丽,田国会.基于关键点序列的人体动作识别[J].机器人,2016,38(2):200-207.]
[6]LIU Changzheng,ZHANG Roughua,GUO Li.Human action recognition algorithm based on localization in complex backgrounds[J].Research and Exploration in Laboratory,2016,35(2):107-113(in Chinese).[刘长征,张荣华,郭理.复杂背景下基于定位的人体动作识别算法[J].实验室研究与探索,2016,35(2):107-113.]
[7]Bobick AF,Davis JW.The recognition of human movement using temporal templates[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,23(3):257-267.
[8]CHENG Haisu,LI Qingwu,QIU Chunchun.Human action recognition algorithm based on improved dense trajectories[J].Computer Engineering,2016,42(8):199-205(in Chinese).[程海粟,李庆武,仇春春.基于改进密集轨迹的人体行为识别算法[J].计算机工程,2016,42(8):199-205.]
[9]LEI Qi,YU Huiping,WU Min.Active semi-supervised affinity propagation clustering algorithm[J].Pattern Recognition and Artificial Intelligence,2015,23(11):961-968(in Chinese).[雷琪,余慧萍,吴敏.主动半监督仿射传播聚类算法[J].模式识别与人工智能,2015,23(11):961-968.]
[10]LI Kun,LIU Peng,LYU Yajie.The parallel algorithms for LIBSVM parameter optimization based on Spark[J].Journal of Nanjing University(Natural Sciences),2016,52(2):343-352(in Chinese).[李坤,刘鹏,吕雅洁.基于Spark的LIBSVM参数优选并行化算法[J].南京大学学报(自然科学),2016,52(2):343-352.]
[11]Pei J,Han J,Mortazavi-Asl B,et al.Mining sequential patterns by pattern-growth:The prefixspan approach[J].IEEE Transactions Knowledge and Data Engineering,2014,16(11):1424-1440.
[12]CHEN Bin,ZHANG Lianhai,NIU Tong.A minimum classification error criterion based linear discriminant analysis method for speech recognition feature[J].ACTA Automatica Sinica,2014,40(6):1208-1215(in Chinese).[陈斌,张连海,牛铜.基于MCE准则的语音识别特征线性判别分析[J].自动化学报,2014,40(6):1208-1215.]
[13]LI Jingyi.Research on human motion recognition based on Kinect depth image[D].Beijing:Beijing University of Posts and Telecommunications,2015(in Chinese).[李靖意.基于Kinect深度图像人体动作识别研究[D].北京:北京邮电大学,2015.]
[14]Jung Hyun-Joo,Hong Ki-Sang.Modeling temporal structure of complex actions using bag-of-sequencelets[J].Pattern Recognition Letters,2017,85(14):21-28.
[15]ZHANG Feiyang,LI Junfeng,SHEN Junmin.Human behavior recognition based on the statistical characteristics of gradient and optical flow[J].Journal of Optoelectronics Laser,2015,26(8):1593-1601(in Chinese).[张飞燕,李俊峰,沈军民.基于梯度和光流统计特性的人体行为识别[J].光电子·激光,2015,26(8):1593-1601.]
[2]Feng Z,Yang B,Li Y,et al.Real time oriented behavior driven 3Dfreehand tracking for direct interaction[J].Pattern Recognition,2013,46(2):590-608.
[3]Yuan Yuan,Zheng Xiangtao,Lu Xiaoqiang.A discriminative representation for human action recognition[J].Pattern Recognition,2016,59(12):88-97.
[4]TIAN Guohui,YIN Jianqin,HAN Xu.A novel human activity recognition method using joint points information[J].ROBOT,2014,36(3):285-292(in Chinese).[田国会,尹建芹,韩旭.一种基于关节点信息的人体行为识别新方法[J].机器人,2014,36(3):285-292.]
[5]YIN Jianqin,LIU Xiaoli,TIAN Guohui.Human action recognition based on the sequence of key points[J].ROBOT,2016,38(2):200-207(in Chinese).[尹建芹,刘小丽,田国会.基于关键点序列的人体动作识别[J].机器人,2016,38(2):200-207.]
[6]LIU Changzheng,ZHANG Roughua,GUO Li.Human action recognition algorithm based on localization in complex backgrounds[J].Research and Exploration in Laboratory,2016,35(2):107-113(in Chinese).[刘长征,张荣华,郭理.复杂背景下基于定位的人体动作识别算法[J].实验室研究与探索,2016,35(2):107-113.]
[7]Bobick AF,Davis JW.The recognition of human movement using temporal templates[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2014,23(3):257-267.
[8]CHENG Haisu,LI Qingwu,QIU Chunchun.Human action recognition algorithm based on improved dense trajectories[J].Computer Engineering,2016,42(8):199-205(in Chinese).[程海粟,李庆武,仇春春.基于改进密集轨迹的人体行为识别算法[J].计算机工程,2016,42(8):199-205.]
[9]LEI Qi,YU Huiping,WU Min.Active semi-supervised affinity propagation clustering algorithm[J].Pattern Recognition and Artificial Intelligence,2015,23(11):961-968(in Chinese).[雷琪,余慧萍,吴敏.主动半监督仿射传播聚类算法[J].模式识别与人工智能,2015,23(11):961-968.]
[10]LI Kun,LIU Peng,LYU Yajie.The parallel algorithms for LIBSVM parameter optimization based on Spark[J].Journal of Nanjing University(Natural Sciences),2016,52(2):343-352(in Chinese).[李坤,刘鹏,吕雅洁.基于Spark的LIBSVM参数优选并行化算法[J].南京大学学报(自然科学),2016,52(2):343-352.]
[11]Pei J,Han J,Mortazavi-Asl B,et al.Mining sequential patterns by pattern-growth:The prefixspan approach[J].IEEE Transactions Knowledge and Data Engineering,2014,16(11):1424-1440.
[12]CHEN Bin,ZHANG Lianhai,NIU Tong.A minimum classification error criterion based linear discriminant analysis method for speech recognition feature[J].ACTA Automatica Sinica,2014,40(6):1208-1215(in Chinese).[陈斌,张连海,牛铜.基于MCE准则的语音识别特征线性判别分析[J].自动化学报,2014,40(6):1208-1215.]
[13]LI Jingyi.Research on human motion recognition based on Kinect depth image[D].Beijing:Beijing University of Posts and Telecommunications,2015(in Chinese).[李靖意.基于Kinect深度图像人体动作识别研究[D].北京:北京邮电大学,2015.]
[14]Jung Hyun-Joo,Hong Ki-Sang.Modeling temporal structure of complex actions using bag-of-sequencelets[J].Pattern Recognition Letters,2017,85(14):21-28.
[15]ZHANG Feiyang,LI Junfeng,SHEN Junmin.Human behavior recognition based on the statistical characteristics of gradient and optical flow[J].Journal of Optoelectronics Laser,2015,26(8):1593-1601(in Chinese).[张飞燕,李俊峰,沈军民.基于梯度和光流统计特性的人体行为识别[J].光电子·激光,2015,26(8):1593-1601.]