Research On Unpiloted Localization Technology Of SLAM Based On The Fusion Of Visual And Inertial Navigation

Driverless vehicle is the trend of automobile in the future.Positioning and navigation is the foundation of driverless technology.The decision-making,planning and control of driverless vehicles need precise positioning information.SLAM(simultaneous localization and mapping)is a new positioning technology for driverless vehicles,which can realize positioning and navigation without prior map information.The traditional navigation technology includes GPS and IMU inertial positioning technology.Both traditional positioning technology and SLAM positioning technology have their own disadvantages and limitations,so the integrated navigation of multiple positioning technologies is the key to achieve high-precision positioning of driverless vehicles.Based on the fusion positioning technology of IMU inertial navigation and visual SLAM,this paper achieves high-precision positioning algorithm of unmanned vehicle.This paper first introduces the basic framework and foundation of pure visual SLAM.Then,we study and analyze the visual odometer as the front of visual SLAM,and propose a new feature matching strategy algorithm to improve the performance of SLAM algorithm.Finally,we study the tight coupling fusion of IMU and visual SLAM.In order to reduce the impact of time stamp deviation of multi-sensor fusion on the system,the time stamp deviation of IMU and camera is calibrated online and real-time.The research of this paper will be based on the open source VI-ORB-SLAM system,and propose the optimized Improved-VI-ORB-SLAM system.The main research work of this paper is as follows:(1).Research on the front-end visual odometer of SLAM.In view of the low efficiency of feature matching strategy in traditional visual odometer,this paper optimizes it from two aspects: feature matching and false matching elimination.(a):The feature matching of traditional SLAM visual odometer adopts BF(Brute-force)violence algorithm.BF matching algorithm has a high computational complexity in the scene with a large number of feature points.To solve this problem,this paper proposes a Hybrid BF-FLANN(H-BF-FLANN)feature matching algorithm based on FLANN(fast library for approximate nearest neighbors)feature matching.The H-BF-FLANN matching algorithm selects BF or FLANN algorithm for rough feature matching according to the number of feature points in the current image frame.Compared with the traditional BF algorithm,the H-BF-FLANN matching algorithm can effectively improve the efficiency of feature matching for different scene images.The Improved-VI-ORB-SLAM system in this paper adopts the H-BF-FLANN matching algorithm for image processing,which can greatly improve the calculation efficiency of the system and reduce the system time consumption.(b): Traditional SLAM uses RANSAC(Random Sample Consensus)algorithm to eliminate feature mismatch.Because RANSAC algorithm randomly selects data for modeling,its calculation efficiency is low.To solve this problem,the Improved-VI-ORB-SLAM system uses the more efficient PROSAC(progressive sample consensus)algorithm to replace RANSAC algorithm.The RANSAC algorithm randomly selects sample points to estimate model parameters.The PROSAC algorithm sorts the data by similarity,and selects the corresponding relationship with high similarity as the subset.In this paper,the Improved-VI-ORB-SLAM system uses PROSAC algorithm to eliminate the mismatches.Compared with the traditional VI-ORB-SLAM algorithm,the calculation efficiency is higher and the accuracy of pose estimation is higher.Theoretical analysis and experimental verification are given later.(2).Research on fusion algorithm of IMU inertial navigation and visual SLAM.Aiming at the problem of time stamp misalignment between IMU and camera,the improved-vi-orb-slam system calibrates the time stamp deviation of IMU and camera in real time online,which can correct the camera feature point through the estimated time deviation to ensure the correct fusion with the pre-integration data of IMU.In addition,the time stamp deviation will be updated as the system optimization variable.Theoretical and experimental results show that the real-time calibration time deviation improves the positioning accuracy and robustness of the improved-vi-orb-slam system.