Research On The Method Of Segmentation And Location Of Overlapping Fruits In Citrus Harvesting Robot

Citrus is one of the fruits widely grown in China,and it is also a fruit with a large amount of trade in the world.At present,citrus picking is mainly based on artificial methods,and there are problems such as seasonality,labor intensity,and high cost.Picking robot can effectively reduce the intensity of manual work and reduce production cost.However,the accurate identification of overlapping fruits in natural environment is one of the problems faced by intelligent picking.Therefore,it is of great significance to study the identification and positioning technology of overlapping fruits for citrus picking robots..Based on the key project of Chongqing key industry common technology innovation,this paper takes the self-developed citrus picking robot ontology as the platform,based on binocular stereo vision technology to carry out the identification of unobstructed overlapping citrus in the natural environment and partial occlusion of overlapping citrus by leaves.Positioning technology research.The main research work of the thesis includes:1.Construction and calibration of binocular stereo vision system.A binocular stereo vision system for the citrus picking robot was built and the binocular camera was calibrated.The results show that the error of the left camera in the direction is 0.29743 and 0.25099 pixels respectively;the error of the right camera in the direction is 0.30286 and 0.25288 pixels,respectively,which meets the accuracy requirements.2.Segmentation and recognition of overlapping citrus.Combining Lab and HSV color spaces,accurate extraction of citrus regions in images is achieved based on color features.In the natural environment,a fruit segmentation method based on concave region simplification and distance analysis was proposed for unconfined overlapping citrus.A fruit segmentation based on contour curvature and distance analysis was proposed for overlapping citrus partially obscured by leaves.The method achieves the effective segmentation of overlapping fruits and combines the least squares ellipse fitting and the circle fitting method to reconstruct the contour.On this basis,the experiments of segmentation and reconstruction of overlapping citrus in outdoor orchard environment were carried out.The three methods of contour reconstruction error,non-coincidence and reconstruction time were used to analyze the segmentation of the proposed method,RANSAC algorithm and Hough transform algorithm.Refactoring effect and efficiency.The results show that the evaluation index based on the method is optimal,which indicates that the fruit segmentation method proposed for the unobstructed and partially obscured overlapping citrus is effective and feasible.3.Three-dimensional spatial positioning of overlapping citrus.The SURF algorithm was used to extract the feature points of the overlapping citrus images.The SSD algorithm,RANSAC algorithm and polar line constraints were used to complete the feature point matching.On this basis,the three-dimensional space coordinates of the overlapping citrus fruit centers were determined.The depth value of the fruit center was used as the basis for judging.The occlusion relationship of overlapping citrus was determined,and the outdoor measurement and contrast experiment was carried out on the depth of overlapping citrus centers.The results show that the average error measured by this method is about 6.38 mm,which meets the requirements of outdoor space positioning.4.Software interface design and implementation of citrus robot binocular stereo vision system.In the Qt5.8.0 software compilation environment,combined with the OpenCV3.10 image processing library,the software interface design of the citrus picking robot vision system was completed,and the visualization of related module functions was realized.