Multi-Objective Recognition And 3D Reconstruction In Apple Picking Environment

In order to solve the problems of high labor costs for fruit picking and picking,and to realize automation and unmanned management of orchard,research on fruit and vegetable picking robots has become a hot topic today.Building the picking environment ahead of time will save time for obstacle avoidance work for robot picking.This paper will mainly use apple fruit trees as the identification and research objects,build fruit tree model and binocular stereo vision system,complete the identification and image segmentation of fruit and foliage on fruit trees,complete the calibration of binocular camera,and determine the calibration conditions suitable for picking environment.To complete the matching and 3D reconstruction of fruit tree images in binocular vision,the main research contents are as follows:By comparing the color space in the visual application,the RGB color space is selected to segment the fruit and the branches in the fruit tree image.A method of image preprocessing is proposed.The image is processed by sharpening and mean filtering.The image is processed by morphology.A binary image is obtained,and a maximum circumscribed circle method is proposed to fit the fruit contour in the binary image,and the image is converted into a two-dimensional pixel point set.The program is programmed to complete the boundary fitting,and the experiment is performed on multiple sets of images.The Hough circular transformation method proves the applicability of the proposed method;the maximum circumscribed rectangle method is proposed to process the image of the foliage,and the recognition of fruit and foliage in the fruit tree image is completed.The calibration of the camera is completed and the basic parameters are obtained.Through the research on the imaging and calibration principle of the camera,the control variable method is used to explore the influence of different calibration plate sizes and different illumination environments on the calibration accuracy of the camera,and the calibration conditions suitable for the picking environment are determined.,to improve the calibration accuracy;finally based on the conclusions obtained,the calibration of thecamera is completed.This paper explores the principle and basic method of binocular stereo matching,compares the traditional binocular matching method,proposes a suitable matching for fruit tree images,obtains depth information,compares with traditional methods,and performs time-consuming experiments;The problem is summarized and a solution is given to reduce the mismatch in binocular matching.The fruit positioning experiment is completed,and compared with the measured coordinates of the laser tracker,the experimental error is analyzed;the depth information acquired after the matching is converted into the point cloud coordinate,and the point cloud coordinates are fitted and weighted after being saved,and the point is The cloud image is colored,the 3D reconstructed image of the final fruit tree is obtained,the point cloud coordinates are acquired in the point cloud processing software,and the similarity and accuracy of the reconstructed result are compared.