Car detection using codebook and Directed Graphical Model

In this paper, we propose a Directed Graphical Model-based car detection method. Cars are represented by codebook, which is generated robust to surface marking. We modeled visual context into boosted MCMC to reduce the effect of background during object detection. Two kinds of spatial context (part-part, object background) and a hierarchical context (part-whole) are used. We incorporate these contexts into a directed graphical model that can provide car detection information in the form of figure-ground segmentation. The inference is conducted using multi-modal Markov Chain Monte Carlo (MCMC) sampling. Experimental results validate the power of the proposed framework for car detection especially in a cluttered environment.