Door Detection without Apriori Color Knowledge

Most existent door detection methods assume that the panels’ colors are known and rely mainly on this apriori knowledge. However, the color is not always available especially when the environment is entirely unknown. In this paper, a door was defined as a rectangular region with almost a homogeneous color and proper size which one can enter or exit through. The proposed door detection framework features no need for any apriori color knowledge. We adopt the determined finite automaton to model the process which starts from motion detection to obtain the potential door regions roughly, then a combined algorithm is used to extract stable and interest edges which bound the subsequent region growing, finally the door regions could be located accurately by fusing all the preceding results. Experiments under various environments show that this framework is effective and adaptive.