Research on Fruit Picking Positioning Based on Binocular Vision

Based on YOLOv5 object detection algorithm, fruit images are extracted from complex backgrounds in response to occlusion of branches and leaves, fruit overlap, and light effects caused by environment. Method of adding center coordinate value of image based on prediction box of fruit image determines centroid coordinate of target image. Then, by optimizing the focus camera and using Zhang Youzheng's calibration method, the image is calibrated using Matlab calibration box. Obtained image is stereo corrected using local stereo matching algorithm to obtain three-dimensional coordinates of centroid of target fruit, which are then converted into physical coordinates for positioning, After actual measurement, a positioning result with an error of ± 20mm was obtained, confirming its applicability for practical positioning operations.