Research On Localization Optimization And Loop Closure Detection Of Unmanned System In Outdoor Environment

SLAM technology applied to outdoor scenes is a hot area in the field of robot research,and is also one of the core technologies in military offensive and defense,search and rescue,exploration and other fields.This paper focuses on the localization optimization and loop closure detection of the unmanned system applied to open outdoor scenes,in which loop closure detection should take into account the mismatching problem of loop closure.The purpose of this study is to figure out the problem that the intelligent unmanned system cannot build a global consistent map with sparse outdoor environment features,which is because of the low localization accuracy caused by the increase of sensor measurement error,more environment interference and noise.In this paper,we design a loop closure detection method based on the fusion of appearance and geo metry information from multi-sensor.This method would provide effective reference information for back-end optimization,and then,the pose and map can be optimized based on the improved graph optimization algorithm from local and global perspectives so t hat we can provide accurate pose and map information for the subsequent navigation module.In view of the UGV model and the sensor model in mathematical modeling,we deduce the pose estimation based on feature points and scan-to-subgraph localization in theory.And then,we compares scan-to-scan localization algorithm with scan-to-subgraph localization algorithm to analyze the reason why we choose to apply the latter to our system.Considering that scan-to-subgraph match is easily into the local optimal solution and the range for matching is wider,this paper provide a solution for a given rough estimation of initial pose based on multi-sensor information fusion.In order to solve the problem of large front-end localization deviation and fast error accumulation,we study the nonlinear optimization theory deeply,and sort out the relationship between key frame pose and constraints.In order to implement an improvement on the graph optimization algorithm,we consider the influence of loop closure constraints emphatically according to the actual application scenarios when we design the objective optimization function.Analyze the causes and influences of the failure of the close-range loop closure detection mechanism in outdoor scenes with sparse features,and d esign the loop closure detection method based on the fusion of appearance and geometry information to deal with this problem.The algorithm extracts the scale invariant ORB features of appearance information,builds the bag of word based on the Kmeans++ algorithm,and store the bag of word with tree data structure.The appearance information forms a description vector according to the bag of word,and combine with the historical description vector to select the candidate frame.RANSAC is used for geometry verification so that we can eliminate the incorrect candidate frame of loop closure,and then,map scan frames and subgraph based on time consistent.After that,the similarity scoring threshold elimination and geometric constraint construction were carried out according to the correlative scan match algorithm,and the branch-and-bound algorithm was used to accelerate the process and improve the real-time performance of the system.According to the above research contents,the design of Gazebo experimental platform,the construction of simulation experimental system and the design of experimental process have been completed successively.Through simulation experiments,the validity of the algorithm has been verified so that the physical experiment foundation has been laid.After that,complete the hardware selection of outdoor physics experiment and the design of the collaborative experiment system of one UAV and one UGV.In the selected outdoor experiment scene,complete the experimental test and data recording according to the designed physical experiment process.Moreover,the applicability,accuracy,real-time performance and robustness of the algorithm are proved qualitatively and quantitatively by comparing and analyzing the experimental data.Sum up all the above work.