Research on Cotton Row Detection Algorithm Based on Binocular Vision

As a relative locating method, machine vision is generally used for automatic navigation of cotton cultivator or cotton insecticide sprayer. However, it is difficult to achieve reliable and stable recognition of crop row with monocular stereo vision system, because it neither can access directly to the depth information of the image, which leads to massive time-consuming calculation, nor possess high-accuracy recognition or a good anti-noise property. This paper presents an algorithm for cotton row detection based on binocular stereo vision to be used for automatic navigation of cotton cultivator. The Zhang's plane calibration is used to obtain the internal and external parameters of the binocular stereo vision. Preprocessing means are applied to distinguish the cotton from soil, stereoscopic match is conducted according to the SIFT operators after the preprocessing of images, after which cotton space three-dimensional coordinates are acquired by parallax distance measuring method, with the elevation information combination of Hough transform, cotton lines are finally detected. The detection results indicate that this method has an accuracy higher than 90%, which primarily meets the need of automatic navigation for cotton cultivator.