Research On Location And Navigation Technology Of Parking Robot Based On Visual SLAM

In recent years,the number of cars has increased year by year,and the difficulty of parking has become more prominent.The application of parking robots in parking lots has become a new and efficient parking solution.However,in indoor or underground parking spaces,external positioning signals such as GPS are missing.The location and navigation of parking robots requires a solution that does not rely on external positioning signals.Visual SLAM is a suitable solution for this scenario.In addition,compared with current laser navigation solutions,vSLAM-based location and navigation technology has the advantages of not requiring the installation of laser reflectors and low cost of sensors,which has great development potential.This paper mainly studies the location and navigation technology of parking robot based on vSLAM.In order to obtain the motion of the robot and local map through the two adjacent images taken by the camera on the robot,this paper first studies the visual odometer based on the ORB feature.First of all,this paper analyzes and compares the feature point extraction algorithms,and through experiments it is concluded that the ORB feature point algorithm consumes significantly less time than the SURF and SIFT feature points.Then this paper compares the performance of the feature point matching algorithms in the case of a large number of feature points,and finds through the experiment that the matching algorithm based on FLANN is more suitable for the application in this case.For the PnP problem in the 3D-2D scene and the ICP problem in the 3D-3D scene,the method o using linear algebra is given first,and then using it as the initial value,the nonlinear optimization method is used to solve.In order to solve the problem of large cumulative error after long-term operation of the visual odometer,the back-end optimization based on the bundle adjustment is studied next.First,this paper analyzes the solution of the general nonlinear least squares problem,and establishes the objective function of the bundle adjustment;then,using the sparsity of the coefficient matrix of the incremental equation,the marginalization method is used to solve the incremental equation.The comparison experiment between the marginalization method and the general method verifies the acceleration effect of the marginalization method on optimization.In addition,the kernel function is also applied in this paper.The paper concludes through experiments that the kernel function can reduce the negative impact of mismatching points on optimization.In order to verify and analyze the location and navigation technology of parking robot based on vSLAM algorithm,a small physical model is finally set up to simulate the operation of the robot in the actual scene.First,the hardware system of the physical model of the robot is built,then the software system of the parking robot is built under the ROS platform,and finally,the simulated parking function of the small physical model is tested and analyzed experimentally.The physical model built in this paper works well in the test experiment.The paper contains 41 pictures,6 tables,and 55 references.