Research On Visual Slam Of Inspection Robot Based On Image Defogging And Semantic Mapping

The patrol robot is used in various indoor and outdoor environments to replace human to perform tasks.Most of the existing patrol robots use laser slam for positioning and mapping,which can complete the work in the face of a single and simple inspection task.However,in the face of complex tasks,there arises a problem that semantic information can not be obtained.In order to solve the above problem,it is necessary to develop a prototype inspection robot with hardware required for visual slam,and to carry out the research on visual simultaneous positioning and semantic mapping technology of the inspection robot,so as to improve its visual positioning tracking and semantic mapping capabilities,and to provide a theoretical basis for improving the intelligent level of the inspection robot.First of all,this paper analyzes the scene characteristics of the inspection robot,determines the hardware requirements of the inspection robot,designs and analyzes the control system and hardware structure system of the inspection robot,and finally develops a prototype of the inspection robot which has the performance parameters,to meet the requirements of the inspection tasks,and which has the hardware basis to achieve the visual slam.It is the basis for the subsequent algorithm research of this paper,and it also provides a platform for the experimental verification.Secondly,in view of the poor robustness of visual slam in foggy environment,this paper studies and improves the image defogging algorithm and proposes an adaptive image defogging algorithm for the unknown environment faced by the inspection robot.The advantages of the adaptive image defogging algorithm are verified by comparative experiments,and the algorithm is used as the front-end of image processing,and the ORBSLAM2 is improved.The comparison results of simulation experiments show that the location and tracking ability of the improved ORBSLAM2 algorithm in foggy environment has been improved.Thirdly,in view of the problems of poor semantic segmentation effect and poor semantic mapping effect when the inspection robot visual SLAM acquires semantic information and semantic maps,the related algorithms is applied and improved,and a semantic segmentation algorithm combined with scene classification is proposed.Through experimental comparison,it is verified that the algorithm can improve the effect of semantic analysis;the semantic mapping technology and semantic segmentation are studied,and ORBSLAM2 is improved by using the semantic segmentation algorithm,realizing the semantic map construction;Simulation and comparison show that the proposed algorithm can improve the effect of semantic mapping.Finally,using the developed inspection robot,the positioning and tracking experiments in foggy environment and semantic mapping experiments in unknown environment are carried out.By comparing the localization and tracking effects of the robot and the effect of the semantic maps,the superiority of the simultaneous localization and semantic mapping algorithm proposed in this paper is verified.