DocumentCode :
1849217
Title :
Free viewpoint action recognition based on self-similarities
Author :
Jiao Wang ; Changhong Chen ; Xiuchang Zhu
Author_Institution :
Jiangsu Provincial Key Lab. of Image Process. & Image Commun., Nanjing Univ. of Posts & Telecommun., Nanjing, China
Volume :
2
fYear :
2012
fDate :
21-25 Oct. 2012
Firstpage :
1131
Lastpage :
1134
Abstract :
Action recognition is an important topic in computer vision and most current work focuses on view-dependent representations. In this paper, we develop a novel free viewpoint action recognition based on Self-similarity matrix (SSM), which tends to be stable across views. We choose Local Self-similarity (LSS) descriptor as our low-level feature, then SSM is calculated by computing the similarity between any pair of frame features. Each video sequence is represented using a diagonal descriptor vector extracted from the SSM. Support Vector Machines (SVM) is employed for classification. The encouraging experimental results on the public IXMAS multi-view data set demonstrate effectiveness of the proposed method.
Keywords :
computer vision; feature extraction; support vector machines; video signal processing; IXMAS multi-view data set; SVM; action recognition; computer vision; diagonal descriptor vector; local self-similarity descriptor; self-similarity matrix; support vector machines; video sequence; action recognition; diagonal feature; local self-similarity descriptor; self-similarity matrix; view independent;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Signal Processing (ICSP), 2012 IEEE 11th International Conference on
Conference_Location :
Beijing
ISSN :
2164-5221
Print_ISBN :
978-1-4673-2196-9
Type :
conf
DOI :
10.1109/ICoSP.2012.6491777
Filename :
6491777
Link To Document :
بازگشت