Discriminative training of patch-based models using joint boosting for occupant classification

This paper presents a vision-based occupant classification method which is essential for developing a system that can intelligently decide when to turn on airbags based on vehicle occupancy. To circumvent intra-class variance, this work considers the empty class as a reference and describes the occupant class by using appearance difference. Context contrast histogram is used to represent the patch appearance. Each class is modelled using a set of locally representative parts called patches that alleviate the mis-classification problem resulting from severe lighting change. The selection and estimating the parameters of the patches are learned through joint boosting by minimizing training error. Experimental results from many videos from a camera deployed on a moving platform demonstrate the effectiveness of the proposed approach.