Guidance lane detection for pesticide spraying robot in cotton fields

Abstract

Natural cotton field images were analyzed in a Lab color space to study the feasibility of lane detection for agricultural robots. The cotton in the images were successfully recognized from the soil background. A maximum variance threshold and median filter were used in preprocessing to obtain binary images and remove noise respectively. Vertical histogram of the images could be divided into right and left regions, and then the guidance points were found from the average positioning information of the two regions. The best guidance lane was located using a Hough transformation which was then used to guide the machine. The parameter identification by this method is efficient and can be used to analyze large numbers of sequential images of the cotton field by an AS-R robot.