Sequential Monte-Carlo techniques and vision-based methods for road signs detection

This paper presents a model-based method for road signs detection and tracking. The object is described by a CAD model and tracked through the sequence. The detection and tracking problem is modeled using the estimation theory of hybrid processes, in which each sign of the database is described as a distinct mode. The proposed state-space modeling is here strongly nonlinear. Hence, one develops a Multiple Hypothesis Particle Filter solution that is based on the theory of the Sequential Monte-Carlo Methods. This solution is then applied in real time to real road image sequences.