Activity recognition based on hidden Markov models

Abstract

Focusing on the application of Intelligent Security Supervisory Control System, this paper proposes a new human activity recognition approach in which the Background Subtraction and the Time-stepping are averaged by weights to implement the precise extraction of moving human contour. In this way, the incompleteness of the extracting objects contour resulting from the color comparability between the human and the background can be resolved. Moreover, an ant colony clustering algorithm is applied to estimate and classify the body posture. Finally, Discrete Hidden Markov Models is used for human posture training, modeling and activity matching to recognize the human motion. Experiment results have shown that this method gives stable performances and good robustness. © Springer-Verlag Berlin Heidelberg 2007.