Recognizing Human Group Behaviors with Multi-group Causalities
Cong Zhang, Xiaokang Yang, Weiyao Lin, Jun Zhu · 2012 IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent Agent Technology · 2012
Human group behaviors are usually composed of several sub-groups. Considering the interaction between groups, this paper presents an algorithm to recognize human group behavior with multi-group causalities. It has two main contributions: (1) we introduce inter-group causality to reflect the interaction between human groups, (2) an improved coding scheme (i.e. Locality-constrained Linear Coding) is used for encoding the causality to go beyond Vector Quantization (VQ). Finally, a simple linear SVM is adopted to learn this model. Our experiment results demonstrate that inter-group causality feature and LLC methods can significantly boost behavior recognition performance.