Feature Point Matching Method of Weak Texture Environment in Farmland Based on Improved GMS-PROSAC Fusion Algorithm

Visual SLAM technology has been widely used in all aspects of life, but has been applied relatively little in the field of weakly textured environments in agricultural fields. To make visual SLAM technology can be better applied in farmland environment, aiming at the problem of low accuracy of direct image recognition and matching in weak texture environment of farmland, A GMS-PROSAC feature point matching algorithm based on ORB feature point extraction is applied. This algorithm integrates the screening principle of the PROSAC algorithm with the smoothness constraint and multi-region generalization construction principle of the GMS algorithm, embedding a polar geometric constraint model that incorporates projection error fusion. It effectively overcomes the impact of light sensitivity and noise interference on matching accuracy. Due to its excessive plant repeatability, a judgment method is proposed to eliminate the wrong matching pairs under correct matching, so that the intelligent farm machine can correctly recognize the complex road conditions in different farmlands for reasonable harvesting. Through experimental verification, the highest matching accuracy can reach 86.0% in the weak texture environment of the farmland, which can be more effective for feature point extraction and matching, that provide new directions for subsequent unmanned agricultural machines that can harvest intelligently in different kinds of fields.