Role of feature position in Recognition of Human Vehicle Interaction

This paper proposes an improvement in the bag of words (bow) model by introducing spatial position of features in interaction representation stage for human vehicle interaction recognition. the spatial positions of features are incorporated along with feature descriptor to obtain the structural information required to correctly classify different kinds of human vehicle interaction. bow lacks structural information as well as the relationship between spatiotemporal features. this renders this bow approach,in its basic form,ineffective for human vehicle interaction recognition. the proposed approach enhances the bow representation by emphasizing the role of spatial feature positions for interaction and incorporates the spatial and temporal relationship between the features. this approach has been tested on the state of the art dataset and achieved accuracy of 92.8%. it has been found that our approach out-performs the bag of words approach and other state of the art work.