Detection Algorithm for Crop Multi-Centerlines Based on Machine Vision

Abstract

The information extracted from crop rows is important in controlling an agricultural robot's travel and operations in a vision-based guidance system. Based on different objectives during the management period of wheat, the study proposes an image algorithm to determine all the centerlines of the targets in a wheat image. The color image was converted into a gray-scale image using the color difference 2G-R-B. A self-adaptation threshold separated the crop rows from the inter crop row spaces. The target regions and points were obtained by analyzing each horizontal scan line in a binary image. The target points were clustered according to the abscissa of the target points of two adjacent scan lines. The centerlines were correctly detected by passing a known point Hough transform (PKPHT). The algorithm requires 0.12 s to determine all the centerlines. Test results showed that different multi-centerlines had been accurately detected in 600 wheat images among 650 images sampled under different natural and field conditions.