Research On The Framework And Matching Algorithm Of AUV Undersea Terrain SLAM

Autonomous Underwater Vehicle(AUV),equipped with multi-beam sonar,inertial navigation system and other equipment,can form the capability of Autonomous close-in and fine exploration of deep sea terrain,and can depict small-scale and more precise information of the earth circle.The accuracy of seabed terrain exploration is closely related to the accuracy of navigation and positioning,which is an important factor to determine the quality of seabed terrain exploration.However,the seabed terrain is covered by water,so it is very difficult to correct the positioning error of the inertial navigation system.Simultaneous Location and Mapping(SLAM)enables AUV to integrate multi-beam sonar bathymetric and inertial navigation location information to construct seabed terrain map,and at the same time correct the accumulated positioning error of inertial navigation system based on the bathymetric matching information.There is a natural connection between the relevant theories and seabed terrain detection in perception,positioning and detection mechanism.The theoretical research on seabed terrain SLAM can explain the coupling mechanism of AUV seabed terrain exploration and AUV autonomous underwater positioning,which is an important theoretical support and technical means for AUV to realize long time sequence and fully autonomous seabed terrain exploration tasks.However,the openness of AUV track in large-scale,borderless deep sea space and the similarity of seabed terrain pose challenges to the robustness of the seabed terrain SLAM algorithm framework and the accuracy of the matching method.Therefore,this paper studies the algorithm framework and matching algorithm of AUV seabed terrain SLAM.Firstly,a parallel search and matching method for seabed terrain matching region was proposed to solve the problems of large scale and borderless seabed terrain environment,high similarity of seabed terrain and no environmental constraint of AUV motion trajectory,resulting in large computation of bathymetric data matching of AUV seabed terrain SLAM and prone to ambiguity and misassociation of matching results.In this method,the accuracy of data association and matching efficiency are improved by constructing a multi-hypothesis submap and using a multi-hypothesis matching method combining the coarse registration with point-to-plane Iterative Closest Point(ICP).In rough registration,the Nearest Neighbor(NN)algorithm of KD tree is parallelized to search the reentrant region of multi-hypothesis submap and prior submap,and the multi-hypothesis that the optimal overlap region of the multiple hypothesis submap and the prior submap is selected by calculating Mean Squared Deviation(MSD).In fine registration,point-to-plane ICP algorithm is used to register the results of coarse registration and output the spatial constraint relation between key frames.Secondly,aiming at the fuzziness and uncertainty of data correlation caused by similarity of seabed terrain and noise of multi-beam sonar,the modeling mechanism of seabed terrain keyframe SLAM factor graph with pose constraint is explained.A framework for Incremental Robust Smoothing SLAM(IRS-SLAM)is proposed.The problem of density of square root factor matrix is solved by periodic reordering,which avoids the increase of algorithm complexity and optimizes AUV pose information incrementally in real time.The invalid closed-loop constraints caused by fuzzy bathymetric information and location uncertainty are removed by periodic local consistency detection and global consistency detection.Finally,the experimental results show that the multi-hypothesis search and matching method of seabed terrain adaptation area can correctly match the bathymetric data in the fuzzy seabed terrain environment,and the i RS-SLAM algorithm framework can effectively identify outliers in the presence of a large number of invalid closed-loop constraints,so that the SLAM optimization results remain accurate.It ensures the positioning and mapping accuracy of AUV in long time series and large-scale seabed environment.