Vision-based obstacle detection and avoidance for the CWRU cutter Autonomous Lawnmower

This paper describes the vision-based obstacle detection system of the CWRU Cutter, an autonomous lawnmower developed for the annual “Institute of Navigation (ION) Autonomous Lawnmower Competition.” Unlike LIDAR sensors commonly found on autonomous vehicles, computer vision systems can provide similar information at drastically reduced prices. Though significantly more cost-effective than LIDAR, these systems have inherent problems due to changing lighting conditions and shadows. This paper investigates the use of image hue and intensity to create a robust, real-time vision-based obstacle detection system for use during the ION competition. Data abstraction methods used to process incoming images for easy combination of information from multiple sensors are also discussed. Using this system, CWRU Cutter correctly identified obstacles in 89% of frames containing fence, 78% of frames containing flowerbeds, and 84% of frames containing boundary lines.