Specific behavior recognition based on behavior ontology

Abstract

In this paper, we propose a novel method based on ontology for human behavior recognition. The state-of-art behavior recognition methods based on low visual features or high visual features have still achieved low recognition accuracy rate in reality. The two most important challenges of this problem are still remaining such as the semantic gap and the variety of appearance of human behaviors in reality. By using prior knowledge, our system could completely detect a behavior without training data of entire process and could be reused in other cases. In experimental results, our method have achieved the encouraging performance on PETS 2006 and PETS 2007 datasets. These results have proved the good prospect of the human behavior recognition system based on behavior ontology.