Research On Navigation Technology Of Greenhouse Tomato-Cucumber Agricultural Robot Based On Machine Vision

The application of machine vision technology to the automated navigation of greenhouse agricultural robots and the realization of autonomous walking of greenhouse agricultural robots is of great importance to the automated operation of greenhouse agricultural machinery.However,greenhouse roads are unstructured and the greenhouse plants have complex structures,which makes it difficult for the automated navigation of greenhouse agricultural robots.In this paper,the greenhouse tomato-cucumber agricultural robot is taken as the research object.The basic research of agricultural robot navigation is done,and a corresponding navigation path fitting system is developed.The main research contents and conclusions of this paper are as follows:(1)Aiming at the problems of small gray value difference and over segmentation when using traditional graying factors to gray the RGB images of cucumber and tomato plants,this paper proposed three new graying factors by analyzing the R,G and B component histograms of cucumber and tomato plant images.Experimental results show that the proposed graying factors can achieve good results in the process of image graying and solve the problems of small differences in gray values and over segmentation.(2)In order to solve the problem of curvature of tomato and cucumber rows,this paper adopted the idea of mathematical differentiation "turning curvature into straightness" to intercept a part of the captured image as the region of interest and solve the problem of crop rows curvature.At the same time,aiming at the tomato and cucumber rows in the fruiting stage,this paper proposes an algorithm to interception the target region based on the vertical projection function of binary image,and the experimental results show that the method can eliminate the interference.(3)According to the distribution characteristics of plants on both sides of rows of greenhouse tomato and cucumber,this paper proposed two algorithms,which are the pixel relative coordinate center method and the cumulative pixel relative coordinate center method.The experimental results show that the accuracy of the cumulative pixel relative coordinate center method is better.(4)Aiming at the shortcomings such as large amount of calculation and unable to meet the real-time requirements of the traditional Hough transform,this paper proposed three navigation path fitting algorithms,which are the initial point Hough transform,the median point Hough transform and the predicted point Hough transform.The experimental results show that the improved initial point Hough transform improves the real-time performance,but its anti-interference performance is low.The median point Hough transform retains the robustness of the traditional Hough transform and greatly improves the real-time performance.The prediction point Hough transform is not only improves in real-time performance,but also more robust than the traditional Hough transform,which can meet the requirements of robot automatic navigation.(5)Based on the characteristics of greenhouse roads,this paper built a hardware platform for the walking mechanism and a vision system of greenhouse agricultural robots.In addition,an image processing system was developed based on Labview2018 and Matlab2019 a,on which real-time image acquisition and real-time fitting of navigation paths were realized.The experimental results show that under the control of the navigation path parameters obtained by the methods proposed in this paper,the maximum deviation between the actual walking path of the experimental platform and the artificially fitted navigation path is 3.5 cm.