Jin Ok Kim, Journal of Internet Computing and Services, Vol. 15, No. 4, pp. 9-20, Aug. 2014
10.7472/jksii.2014.15.4.09, Full Text:  HTML
Keywords: Action sequence detection, Categorical action classification, Bio-inspired Action perception, Unsupervised learning

Abstract

Making good predictions about the outcome of one's actions would seem to be essential in the context of social interaction and decision-making. This paper proposes a computational model for learning articulated motion patterns for action recognition, which mimics biological-inspired visual perception processing of human brain. Developed model of cortical architecture for the unsupervised learning of motion sequence, builds upon neurophysiological knowledge about the cortical sites such as IT, MT, STS and specific neuronal representation which contribute to articulated motion perception. Experiments show how the model automatically selects significant motion patterns as well as meaningful static snapshot categories from continuous video input. Such key poses correspond to articulated postures which are utilized in probing the trained network to impose implied motion perception from static views. We also present how sequence selective representations are learned in STS by fusing snapshot and motion input and how learned feedback connections enable making predictions about future input sequence. Network simulations demonstrate the computational capacity of the proposed model for motion recognition.

Statistics
Cite this article