Research On Robot Disassembly Target Reconstruction Technology Based On Machine Vision

With the rapid development of China’s nuclear industry,the demand for nuclear safety devices and technologies is becoming increasingly urgent.This paper,supported a national key research and development project,focuses on the key technology of disassembly operation target reconstruction in the development of nuclear emergency robots.The main research contents and achievements are as follows:(1)Based on the tele-operated nuclear emergency disposal robot,a general plan for disassembly operation target reconstruction was proposed based on the analysis of nuclear disassembly operation scenarios and object characteristics.The disassembly target reconstruction process was established.The design of the disassembly target object collection system based on depth cameras was completed,the type of depth camera was determined,and the design of radiation protection for vision sensors was completed.The collection space analysis and collection path design were completed based on the analysis of the robot’s reachable operation space,and the image collection operation range was established.(2)Noise characteristics of nuclear radiation images were analyzed.Using priori denoising methods and various methods such as peak signalto-noise ratio and structural similarity index quantitative analysis,it was verified that nuclear radiation image noise also belongs to additive noise.In combination with the application requirements of the embedded platform of the nuclear disassembly robot,a Res-CNN neural network image denoising algorithm was proposed.An image denoising dataset was built,including 2960 training sets,370 validation sets and test sets.An evaluation standard for nuclear noise image denoising based on peak signal-to-noise ratio(PSNR)and structural similarity(SSIM)index was adopted.The effectiveness of the algorithm was verified through image denoising experiments and analysis.(3)A 3D reconstruction of the disassembly target was carried out using the ICP iterative point cloud registration algorithm with global and local feature fusion and the Poisson surface reconstruction method based on region growing.In dealing with the problems of point cloud noise and redundancy,the data volume was effectively reduced while maintaining the details of the point cloud,and the efficiency of point cloud registration was improved through the farthest point sampling and voxel network downsampling algorithm.For the geometric shape reconstruction of 3D point cloud data of disassembly operation targets,the proposed Poisson surface reconstruction method based on region growing not only improved the computational efficiency and local accuracy but also effectively reduced the 3D reconstruction holes caused by sparse point clouds.(4)The 3D reconstruction experiment of the disassembly operation target was completed on the disassembly robot experimental platform.The visual information of the disassembly objects was obtained through the disassembly operation target collection experiment.The 3D reconstruction of the disassembly target simulation body was realized using the 3D reconstruction method proposed in this paper,verifying the effectiveness of the 3D reconstruction method of disassembly target in this paper.The research results of this paper laid the foundation for robots to carry out precise and efficient disassembly operations in complex environments.