| They are easily corroded and damaged by long-term seawater erosion for pipelines in the seawater system of nuclear power plants,so they need to be regularly inspected and maintained.Aiming at the positioning requirements of the pipeline crawling robot,Robotic systems need provide accurate pose information for the video detection and foreign object grasping of the robot in the pipeline,which is convenient for the robot to continue to detect,polish and spray.Taking the binocular fisheye camera as the main sensor,the pipeline crawling carrier system was built as the experimental platform,and the principle exploration,optimization or improvement,and experimental analysis of the binocular depth estimation and visual inertial SLAM algorithms were carried out.There are a T-tube and bend tube combination and an empty corridor were selected for experimental testing during the experiment.The specific content is summarized as follows:(1)Process the information obtained by the vision sensor.The imaging principle,parameter calibration and methods involved in image acquisition and image preprocessing of the camera are introduced in detail,and the algorithm research is carried out for the distortion correction of fisheye images.so as to extract a large field of view for the features in the subsequent pipeline environment and indirectly compensate for the disadvantages of weak texture.(2)Target ranging and reconstruction according to the environment inside the pipeline.In the form of local feature detection and regional growth mode growth,the internal target image of the pipeline is extracted and matched,and the false matching point is rejected.One-dimensional search and two-way matching are used to realize the densification of sparse feature points.and the distance estimation and three-dimensional scene reconstruction of objects in the pipeline are realized by dense parallax.(3)Visual and IMU information fusion processing analysis.Aiming at the insufficient positioning performance of pure vision SLAM in the weak texture area,the IMU and visual information are fused,and the fisheye model is added to construct the fisheye visual inertial system.The visual inertial SLAM of pure vision SLAM,visual and IMU fusion SLAM,and visual inertial SLAM added to fisheye model are verified by dataset experiments,and the experimental results show that the addition of IMU improves the overall performance of the system by more than 20%.the addition of fisheye model has no obvious difference in visual inertial SLAM performance.Therefore,it provides a basis for the subsequent real-world environment to exclude the interference of the fisheye model on the system.(4)Fisheye visual inertial system real-world experimental analysis.We build a simple experimental platform and experimental environment.Compare experiments to verify the perception ability of the system in the environment with less texture by detect feature points in wall and ground environments.We also use it for local experimental testing in the real environment of the pipeline to verify the effectiveness and the corridor to verify the robustness of the system at long distances.Experimental results show that the system can meet the basic requirements of positioning in pipelines. |
