Study On Stereo Vision Navigation Method Based On Invariant Feature Extraction Of Green Crops

The farmland environment is complex and changeable.The stereo vision-based navigation method has certain requirements on the amount of information in image,otherwise the crop will be crushed by agricultural robot.The feature information of the image describes a certain property of the image,and the feature points provide the basis for techniques such as feature matching and three-dimensional calculation.Therefore,based on maintaining the green crop characteristic scale,illumination and rotation invariance,it is our goal in this thesis to improve the number of feature points and increase the number of feature points matching.In this thesis,the principle of SIFT feature points extraction algorithm is firstly analyzed,and the detailed experimental results are given for each process of the algorithm.In order to increase the number of feature points used for feature matching,the image is enhanced by embedded Gaussian homomorphic filtering into the process of constructing the Gaussian pyramid.However,the image is darker and the detail information is not obvious.Therefore,the histogram equalization is used to enhance the image.In multi-scale images effectively are improved the problem of scale and illumination by the Gaussian homomorphic filtering and the histogram equalization,and increases the number of feature points by at least 2 times.It is proved by experiment results that the SIFT algorithm embedded by Gaussian homomorphic filtering maintains the scale,illumination and rotation invariance of feature points while increasing the number of feature points.After that,three-dimensional information of crop is obtained based on the feature points matching by using parallax,and an elevation image based on three-dimensional information is constructed.Secondly,the gray contour,binarization,filtering,edge extraction operator and morphological methods are used to extract the whole edge contour of the green crop.The obtained elevation image and edge contour image are further fused to generate a confidence density image.Finally,based on the confidence density image,the navigation model is constructed,and five kinds of operational state equations for controlling agricultural robots are designed by the constraints of navigation angle and side distances.The S-type and O-type paths are used to verify the operational equations of agricultural robots.The experimental results show that the lateral deviation of agricultural robots is [-8,10]cm.At the same time,the navigation angle of the S-type path and the angle actually rotated of the agricultural robot are fitted,and the angle of the navigation angle(?)and the actual rotation angle of the agricultural robot(σ)satisfy the Fourier function with the correlation coefficient being 0.96.The function is verified by S-type path,and the obtained data of the actual rotation angle of the agricultural robot(σ)are verified by the O-type,showing that the mean error and variance are 0.7° and 1.5°,respectively.