Recently, we are developing a vision-based traffic surveillance system (TSS) on the Internet. The first priority for vision-based TSS, is its ability to segment a road scene. In this paper, we propose an effective segmentation technique for road traffic scenes. Since road scenes are usually degraded, it is difficult to obtain accurate segmentation results. For higher quality result a variety of features such as color, blurring, and noise need to be considered. Images are modeled using the Markov random field (MRF) model, which is resistant to degradation. We use a genetic algorithm (GA) as optimization algorithm, which can effectively deal with combinatorial search problems. The experimental results show that the proposed method is effective at segmenting real image and has application possibility in automatic vehicle extraction system.
展开▼