Environment mapping with stereo vision and Belief Propagation algorithm

Creating a top view map of an environment has been one of the most important issues that navigation systems are dealing with. In this paper we discuss this subject and detecting the scene's most important obstacles. We obtain information from the scene by moving the stereo camera and photographing from different views and by this work we create a stereo dataset from our environment. We choose an appropriate matching algorithm and by that, we extract the dense disparity map for each frame of images. Having the disparity map and depth map allows creating a field of coordination that consists of obstacles placement and they are attained by applying triangulation calculations on the depth map. So the local map of the environment will be created based on this coordination. By photographing the scene several times and incorporating the outputs by the help of Bayes Theorem we try to decrease the destructive effects caused by the environment's dynamism. At the end, combining these local maps and applying our optimization algorithm on them which is Belief Propagation algorithm leads to a global map with reduced amount of errors. BP is one of the algorithms used in optimizing marginal probabilities and acts very well on graphical models like MRF, therefore we chose it for our optimization problem since it fits very well in our matter.