Improving Localization Performance of Robot Using Obstacle Recognition and Probability Model through Image Processing

Abstract

In this paper, we propose an effective localization method with only a stereo camera that has obstacles using particle filter. When localization with flow planning rather than robot scanned map, the error of localization increases when there is an obstacle. To solve this problem, First, we propose two types of obstacle recognition method: "Image Split Obstacle" and "Obstacle In Image" through image processing using the Opencv contour function. Afterwards, we solve the problems caused by the particle filter weight calculation process through a new sensing model using interval angle. In addition, we propose two probability models that can solve the problem of inconsistency between the number of landmarks of robots and particles. After that, we suggest an effective robot localization method by presenting a probability model that considers obstacles. As a result, the probability model considering obstacles showed an error rate reduction of about 45% compared to the existing model that does not considering obstacles.