Research On Simultaneous Localization And Mapping Method Of Multi-sensor Fusion Underwater Robot Based On Sonar

Accurate and reliable navigation technology is the fundamental guarantee for underwater robots to accomplish various underwater tasks.Due to the special characteristics of the underwater environment,traditional navigation technology research cannot satisfy the underwater robots to perform underwater tasks.Simultaneous localization and mapping(SLAM),as an emerging navigation technology,is becoming one of the key technologies to achieve autonomous navigation for underwater robots due to its high compatibility with sensor types.However,compared with the laser and vision SLAM solutions used in land robots,the underwater environment faces problems of light attenuation and scattering,uneven illumination or low visibility,and traditional localization and mapping methods face great challenges.And GPS signals cannot penetrate the water surface to provide accurate positional reference for the carrier.Acoustics,as the main means of underwater perception,plays an important role in the field of underwater positioning and navigation,and environmental detection.Therefore,it is important to study how to combine the physical characteristics and information acquisition mechanisms of acoustics and vision to obtain relatively accurate underwater environmental information and further construct underwater maps to provide position reference for underwater robots to further realize autonomous navigation and positioning of underwater robots.In this thesis,we propose various underwater simultaneous localization and map building methods with sonar-based sensors,and try to improve and simultaneously combine the visual SLAM framework in the SLAM framework with particle filtering and graph optimization as the back-end optimization for the research of underwater SLAM framework.The main works are as follows:(1)Processing the publicly available sonar datasets of submarine caves and converting them into 2D laser-like data form for encapsulation to produce the datasets.For the long scanning period of sonar sensors,the use of IMU and odometer is introduced to interpolate the sonar point cloud data for motion distortion removal.In view of the low reliability of ambiguous sonar data in the underwater environment,radius filtering is introduced to process the sonar point cloud data to be scanned and matched,which reduces the mis-matching rate of subsequent scanning and matching.(2)The processed sonar dataset is tested for localization and map building using particle filtering as the back-end optimized laser SLAM framework,and is improved for the long sonar scan period and low sonar data reliability by adding a distance limit for its scan matching trigger condition.The maximum expectation estimation algorithm is introduced in the scan matching to make the position estimation obtained from the scan matching more accurate.Finally,the data are compared using the laser SLAM framework with graph optimization as the back-end optimization,and the experimental results are obtained that the map building and localization effect of the improved particle filtering framework can basically reach the map building and localization effect of the graph optimization framework in less time.(3)A multi-sensor fusion SLAM framework is investigated,and a submerged SLAM framework for sonar,vision and IMU fusion is proposed using processed sonar scan data and camera image data,and the visual data matching error is normalized and fused with the sonar data matching error to obtain a better positional estimation by minimizing the joint error.