Obstacle Detection and Avoidance For Mobile Robots Using Monocular Vision

摘要

This paper proposes a robust approach for obstacle detection and avoidance algorithm using a single camera. Monocular Vision using single camera architecture cannot identify depth with a single image and thus depends on pixel gradient or keypoint extractors to identify traversable path and obstacles. Pixel gradient does not work well where there are shadows and sharp illumination changes and keypoint extractor does not work well in the absence of dense texture. In this paper we propose an algorithm that is able to use edges as keypoints along with pixel gradient. The entire algorithm was successfully tested on Sphero RVR Rover platform that uses Raspberry Pi and a color camera with IR. The proposed method performs well in obstacle detection and obstacle avoidance and is potentially an alternative to a binocular solution.