Towards obstacle Identification by Markovian-Evolutionary Segmentation

When a robot navigates autonomously through an environment, it is necessary that it can accurately identify the obstacles with which the robot can collide. To specifically estimate the time of contact with the obstacle, the segmentation of the obstacle must be as accurate and rapid as possible. This paper presents a work in progress where we show a proposal to segment obstacles using probabilistic Hidden Markov Chains with which the search for the best position of control points is optimized by means of an evolutionary algorithm. Although it is a work in progress, the use of a Genetic Algorithm allows not to explore all possible contour configurations, while the Hidden Markov Chain allows a balance between a good positioning of the control points on the object and a good distribution of the points on the contour.