The Optimization of Localization and Navigation for Vision-based Robot

This paper studies the localization and navigation of vision-based robots. In the first step, obtaining the video stream by the monocular camera to get high-quality image information. Then we do a series of preprocessing, such as noise reduction, space conversion, contour extraction, to extract color and shape information. In the second step, getting edge information using a corner detection algorithm combined with weight assignment and constraint, acquiring the current position using the multi-region multi-threshold edge localization algorithm. In particular, for the multi-object locate problem caused by the change of the robot viewing angle, using multi-image stitching matching algorithm to locate, which based on the spatial intermediate point. The third step, we design a number of basic action groups, robot will select the appropriate action group according to the current position and obstacle information, so the robot can pass quickly by bypassing, rolling or crossing over.