Autonomous Navigation And Row Detection in Crop Fields Using Computer Vision
Abstract
This thesis documents the development of an autonomous row crop guidance system for a differentially wheeled agricultural robot. Computer vision is used to identify rows and estimate their position and orientation. The estimates are used as measurements in a Kalman filter, before being supplied to a row controller which attempts to make the robot follow the row closely. A simulator with a realistic robot and fields have been developed, and all methods were tested on real robots. Tests have been done on an indoor recreation of a field and outdoors on a real carrot field. The result is a system which is able to autonomously traverse an entire field in simulations and recreated fields.