Intersection Detection and Navigation for an Autonomous Greenhouse Sprayer using Machine Vision

Abstract

Development of a robotic greenhouse sprayer can provide more accurate spraying ofpesticides and fungicides, reduce operational costs, and decrease health risks associated withhuman exposure to dangerous chemicals. A visual navigation system capable of tracking a path,detecting intersections, and navigating through intersections was developed for a six-wheeldifferential steering vehicle.<br><br>Path navigation was accomplished using digital images taken from a single CCD camera mountedon a robotic greenhouse sprayer. Intersection detection was accomplished by first classifying pixelsof an image as path or non-path. Left and right path edges were determined from the thresholdimage using least squares fitting. Path pixels extending beyond a specified distance from the left andright edges indicated an approaching intersection. The distance from the vehicle to the intersectionwas estimated based on the set position and orientation of the camera using projective geometry.Path following was carried out by reducing path error between the vehicle center and path centerusing a PID controller. Intersection navigation was accomplished by 1) tracking ground features toguide the vehicle to a desired position in the intersection, 2) proceeding with a turn, and 3)concluding when the vehicle was aligned with the next path. Testing verified the vehicles ability tofollow a path, detect intersections, and navigate intersections. A more detailed report of this study isavailable from Younse (2005).