Method of Localization of Racks with Biomaterial for Robot Grasp Based on Segmented Contour Processing

This paper is devoted to the solution of the problem of moving a rack with test tubes filled with blood to the area of aliquotation operation is considered. To solve this problem, a vision system based on the YOLOv8 neural network was used and a gripper was designed. The neural network is trained to perform image segmentation, the purpose of which is to determine the contour of the rack. To correct the resulting image segmentation defects, an algorithm has been developed that allows to create an accurate bounding frame and rack orientation. A software package has been developed in Python, which includes modules for image segmentation by a neural network, an algorithm for determining the contour and orientation of a rack, and a robot control module.