| Background modeling is widely used in many fields, such as intelligent surveillance,robot vision, and video conference. Although background subtraction method has manyadvantages, such as moderate algorithm complexity, clear physical meaning, and obtainingimportant information directly. However, we still need to face a lot of problem to solve, suchas the “ghost”, the “stationary object” and the illumination sudden changes due to thecomplexity of the scene, different kinds of the cameras and the imperfection of the algorithm.The main work of this thesis is as follows:1. The methods for tackling the “ghost”,“stationary object” and the fast illuminationchange are summarized. Several popular adaptive dynamic background modeling methods areintroduced and analyzed.2. A dual-direction background modeling method is implemented in a new way to detectghosts and stationary foreground. The dual-direction background model is constructed byforward and backward modeling respectively, and which can detect not only the movingforeground but also the stationary foreground and the ghosts occur when objects start or stopmoving. Finally, the static foreground can be detected by getting rid of ghost and cavitycombined with the edge similarity of the dual-direction difference results. Extensiveexperiment results demonstrate that the proposed algorithm is effective and efficient ineliminating ghosts and detecting stationary foreground.3. An outdoor illumination change model is studied. Based on this model, a chromaticitydifference model and a brightness ratio model and intensity ratio are used to separate detectedforeground pixels into moving object pixels and false foreground pixels. In order to improvethe efficacy of the algorithm, the prejudge of sudden illumination change and the background updating for pixels with sudden illumination change are incorporated into the algorithm. Theexperimental results show this method has an excellent performance under variousillumination changing conditions. |
PDF Full Text Request