基于SIFT和HOG特征融合的人体行为识别方法
|
摘要
行为识别是视频分析的一个核心任务,而行为特征的提取与选择直接影响识别效果。针对单一特征往往受到人体外观、环境、摄像机设置等因素的影响而识别效果不佳的问题,提出一种分别提取尺度不变特征变换(SIFT)和方向梯度直方图(HOG)的特征并形成融合特征,再利用支持向量机(SVM)完成特征分类的行为识别方法。基于Matlab人体行为识别和检测的研究,通过采用KTH和Weizmann人体行为库来验证该算法的有效性。实验结果表明,该算法在人体行为识别中识别率可达到90%以上,比单独使用上述两种特征或者其他传统的描述子更高效,同时也能更好地适应光照等外部因素的变化,得到更好的识别率。
Action recognition is a core task of video analysis,and the extraction and selection of behavior features directly affect the recognition effect. In view of the problem of bad recognition effect for single feature caused by human appearance,environment,camera settings and other factors,we propose a action recognition method which respectively extracts scale-invariant feature transform(SIFT) and direction gradient histogram(HOG) and fuse them,and then adopts support vector machine(SVM) for feature classification. Based on research on Matlab human action recognition and detection,the validity of the algorithm is verified by KTH and Weizmann human behavior libraries. Experiment shows that the recognition rate of the algorithm in human action recognition can reach more than 90%,which is more efficient than above two features alone or other traditional descriptors,and can better adapt to the change of external factors such as illumination and get better recognition rate.
Action recognition is a core task of video analysis,and the extraction and selection of behavior features directly affect the recognition effect. In view of the problem of bad recognition effect for single feature caused by human appearance,environment,camera settings and other factors,we propose a action recognition method which respectively extracts scale-invariant feature transform(SIFT) and direction gradient histogram(HOG) and fuse them,and then adopts support vector machine(SVM) for feature classification. Based on research on Matlab human action recognition and detection,the validity of the algorithm is verified by KTH and Weizmann human behavior libraries. Experiment shows that the recognition rate of the algorithm in human action recognition can reach more than 90%,which is more efficient than above two features alone or other traditional descriptors,and can better adapt to the change of external factors such as illumination and get better recognition rate.
引文
[1] 凌志刚,赵春晖,梁彦,等.基于视觉的人行为理解综述[J].计算机应用研究,2008,25(9):2570-2578.
[2] 周霞,柳絮青,王宪,等.基于特征融合的人体行为识别算法[J].计算机工程与应用,2013,49(7):162-166.
[3] 梁琛华,常青,胡谋法.基于加权Hu矩和 HOG特征的自适应融合人体行为识别新方法[J].现代电子技术,2014,37(1):14-18.
[4] 崔广才,窦凤平,王春才,等.基于傅里叶与局部特征结合的人体姿态识别方法研究[J].长春理工大学学报:自然科学版,2016,39(1):82-87.
[5] ZHANG Jiatao,TSOI A C,LO S L.Scale invariant feature transform flow trajectory approach with applications to human action recognition[C]//International joint conference on neural networks.Beijing,China:IEEE,2014:1197-1204.
[6] 李新德,刘苗苗,徐叶帆,等.一种基于2D和3D SIFT特征级融合的一般物体识别算法[J].电子学报,2015,43(11):2277-2283.
[7] OJALA T,PIETIKAINEN M,HARWOODD.Performance evaluation of texture measures with classification based on Kullback discrimination of distributions[C]//Proceedings of 12th international conference on pattern recognition.Jerusa- lem,Israel:IEEE,1994:582-585.
[8] DALAL N,TRIGGS B.Histograms of oriented gradients for human detection[C]//IEEE computer society conference on computer vision and pattern recognition.San Diego,CA,USA:IEEE,2005:886-893.
[9] POPPE R.A survey on vision-based human action reco-gnition[J].Image and Vision Computing,2010,28(6):976-990.
[10] SIVIC J,ZISSERMAN A.Video Google:a text retrieval approach to object matching in videos[C]//Ninth IEEE international conference on computer vision.[s.l.]:IEEE,2003:1470-1477.
[11] ZHU Lijian,ZHOU Qiaodi,LI Zhu.A new method of feature description for human action recognition[C]//8th international conference on intelligent human-machine systems and cybernetics.Hangzhou,China:IEEE,2016:396-400.
[12] 朱道广.基于视觉词袋模型的图像分类研究[D].郑州:解放军信息工程大学,2013.
[13] SHAO Ling,MATTIVI R.Feature detector and descriptor evaluation in human action recognition[C]//ACM international conference on image and video retrieval.Xi’an,China:ACM,2010:477-484.
[14] THURAU C,HLAVAC V.Pose primitive based human action recognition in videos or still images[C]//IEEE conference on computer vision and pattern recognition.Anchorage,Alaska,USA:IEEE,2008:1-8.
[15] SCHULDT C,LAPTEV I,CAPUTO B.Recognizing human actions:a local SVM approach[C]//Proceedings of the 17th international conference on pattern recognition.Cambridge,UK:IEEE,2004:32-36.
[16] HO W T,HAO W L,YONG H T.Two-stage license plate detection using gentle Adaboost and SIFT-SVM[C]//First Asian conference on intelligent information and database systems.Dong Hoi,Vietnam:IEEE,2009:109-114.
[2] 周霞,柳絮青,王宪,等.基于特征融合的人体行为识别算法[J].计算机工程与应用,2013,49(7):162-166.
[3] 梁琛华,常青,胡谋法.基于加权Hu矩和 HOG特征的自适应融合人体行为识别新方法[J].现代电子技术,2014,37(1):14-18.
[4] 崔广才,窦凤平,王春才,等.基于傅里叶与局部特征结合的人体姿态识别方法研究[J].长春理工大学学报:自然科学版,2016,39(1):82-87.
[5] ZHANG Jiatao,TSOI A C,LO S L.Scale invariant feature transform flow trajectory approach with applications to human action recognition[C]//International joint conference on neural networks.Beijing,China:IEEE,2014:1197-1204.
[6] 李新德,刘苗苗,徐叶帆,等.一种基于2D和3D SIFT特征级融合的一般物体识别算法[J].电子学报,2015,43(11):2277-2283.
[7] OJALA T,PIETIKAINEN M,HARWOODD.Performance evaluation of texture measures with classification based on Kullback discrimination of distributions[C]//Proceedings of 12th international conference on pattern recognition.Jerusa- lem,Israel:IEEE,1994:582-585.
[8] DALAL N,TRIGGS B.Histograms of oriented gradients for human detection[C]//IEEE computer society conference on computer vision and pattern recognition.San Diego,CA,USA:IEEE,2005:886-893.
[9] POPPE R.A survey on vision-based human action reco-gnition[J].Image and Vision Computing,2010,28(6):976-990.
[10] SIVIC J,ZISSERMAN A.Video Google:a text retrieval approach to object matching in videos[C]//Ninth IEEE international conference on computer vision.[s.l.]:IEEE,2003:1470-1477.
[11] ZHU Lijian,ZHOU Qiaodi,LI Zhu.A new method of feature description for human action recognition[C]//8th international conference on intelligent human-machine systems and cybernetics.Hangzhou,China:IEEE,2016:396-400.
[12] 朱道广.基于视觉词袋模型的图像分类研究[D].郑州:解放军信息工程大学,2013.
[13] SHAO Ling,MATTIVI R.Feature detector and descriptor evaluation in human action recognition[C]//ACM international conference on image and video retrieval.Xi’an,China:ACM,2010:477-484.
[14] THURAU C,HLAVAC V.Pose primitive based human action recognition in videos or still images[C]//IEEE conference on computer vision and pattern recognition.Anchorage,Alaska,USA:IEEE,2008:1-8.
[15] SCHULDT C,LAPTEV I,CAPUTO B.Recognizing human actions:a local SVM approach[C]//Proceedings of the 17th international conference on pattern recognition.Cambridge,UK:IEEE,2004:32-36.
[16] HO W T,HAO W L,YONG H T.Two-stage license plate detection using gentle Adaboost and SIFT-SVM[C]//First Asian conference on intelligent information and database systems.Dong Hoi,Vietnam:IEEE,2009:109-114.