Occupant classification for smart airbag using Bayesian filtering
Shih-Shinh Huang, Pei‐Yung Hsiao · 2010
Occupant classification is essential for developing a smart airbag system that can intelligently decide to either turn off or deploy according to the type of the occupants. This paper presents a probabilistic approach to recognize the occupant type from a video sequence. Instead of assuming that the frames are mutually independent, we take the relation between two consecutive frames into consideration. Thus, the problem of occupant classification is formulated by introducing the Bayesian filtering which imposes both transition and measurement terms for the inference of the occupant class. For evaluating measurement term, the higher-order Tchebichef moments of edge maps is computed and then an Adaboost learning algorithm is applied to select a set of discriminative moments as the features. For incorporating the temporal coherence, a finite state machine is used to model the transition probabilities among the occupant classes. Finally, the occupant type is estimated by maximizing the posterior probability. Experimental results for several videos with illumination variation are provided to validate the proposed approach.