Road Image Segmentation Markov Chain Monte Carlo Method

Road segmentation is the basis and key problem in visual navigation of mobile robots. The actual road is often divided into structured road and unstructured road. Structured road generally refers to the better-structured highway that has clear lane line and obvious boundary. Meanwhile, unstructured road generally refers to the worse-structured road that does not have clear lane line and obvious boundary. Due to the the complexity of the road environment, when using the traditional image segmentation method, causing pixels overlap or adhesion, caused image the partitioning of the error rate is higher. Therefore, in order to solve the above problems, we have adopted the following method to investigate:(1)We do some study on Markov chain Monte carlo (MCMC) method. We introduce the origins of MCMC, Markov chain Monte Carlo simulation, the basic principle of MCMC. Also, we describe two MCMC methods that is commonly used——Gibbs sampling method and Metropolis-Hastings algorithm. Then, we analyze the convergence of MCMC method.(2)Apply the MCMC (Markov chain Monte carlo) method for road segmentation. The basic principle is:first formulaic the image segmentation problem, which is conducive to cut up the road image directly and effectively. Then we establish the road image models,which can make the image segmentation problem more accurate. Due to the convergence of Markov chain, compared to the traditional road segmentation method, the MCMC results has smaller error and better convergence.(3)However, in order to make the effect of road segmentation to have further improved, we went on to introduce the GA(Genetic Algorithm) which causes MCMC to MCMC-GA. Due to the cross, mutation and selection operations of GA, the Markov chains become more and more stable, so MCMC-GA makes the road segmentation more accurate. Meanwhile, compared to MCMC, MCMC-GA has less errors and more accurate for segmentation results.