Vision-Based Obstacle Avoidance of Mobile Robot Using Quantized Spatial Model

Abstract

Problem statement: Problem of moving a robot through unknown environment has attracted much attention over past two decades. Such problems have several difficulties and complexities that are unobserved, besides the ambiguity of how this can be achieved since a robot may encounter obstacles of all forms that must be bypassed in an intelligent manner. This research had been aimed to develop a system that was able to detect obstacles in a mobile robot's path using a single camera as only sensory input and to achieve the target point in optimized manner. For this reason, algorithm which took total path length and safety into account was developed. Approach: To control movement of robot from a starting to a target point inside the site where obstacles can obstruct the way of robot, real-time software-specially tailored for this purpose-was necessary to develop. To analyze and to process scene images captured by a (vision) camera, camera was installed at the top over the center of site in a way that it covered whole site through which sufficient image information could be delivered. From sequentially captured images that was manipulated through image processing and computer vision, the system built a representative site model, whose ingredients were gridded squares as a result of quantized spatial plane of site and then it began planning the desired routing path. Results: For building a robot path, less computing effort was necessary because grid information was much easier to deal with than pixels and only a minimum amount of stored data of symbolic site model from current and previous state was necessary. Conclusion: Using a quantized spatial domain, a less computational effort was necessary to control movement of robot with the ability of obstacle detection and avoidance.