New algorithm for machine vision navigation of farm machine based on well-ordered set and crop row structure

A basic requirement of automatic navigational system for farm machine is to recognize precisely the crop rows in various field conditions. In this paper, a new approach based on the density differences between weeds and plants was proposed to detect straight row lines, in which the subsets of the plant pixels were used to estimate the centers of crop rows. The green contents of the row images were chosen to eliminate the background soil features relatively completely compared with the Excess Green. The navigational parameters were obtained through fitting the row center dots. For nonlinear projection region L∈［Height/3,2*Height/3］ of the camera where projecting errors can be neglected, a navigation system equation was established at a platform speed of 1 m/s and sampling interval of 0.1 s. Experimental results indicated that the tracking error of the navigation system was reduced with an angle error about 1° and position error about 1 mm. The system could work well under a wide range of illumination situations and was insensitive to small area weeds.