Research On Real-time Multi-camera SLAM System

In recent years,Simultaneous Localization and Mapping(SLAM)technology has developed rapidly,and related research results have been widely used in many fields such as autonomous driving,robot autonomous navigation and smart home.As a key technology of mobile robot autonomous localization,SLAM technology uses sensors mounted on the robot to obtain information about the unknown environment,estimate the robot’s own position and pose,and construct a map of the working environment.The early laser SLAM technology mainly relies on laser sensors to obtain point cloud data to realize unknown environment perception and mapping.However,lidar has the problems of large body size and high cost.Researchers increasingly favor visual sensors that are light,cheap and rich in information,making visual SLAM gradually become a research hotspot.Facing the actual complex environment,the current visual SLAM technology still has the problem of poor anti-interference ability.It is mainly reflected in the weak texture environment and illumination change will lead to the failure of visual SLAM algorithm.Multi-camera visual SLAM system uses multiple cameras as sensors,which can obtain visual information of a large field of view,so as to improve the influence of weak textures and illumination changes on the algorithm.Even if one camera is occluded,the remaining cameras can still work normally.Therefore,multicamera visual SLAM has better robustness.However,the multi-camera SLAM system needs to process multiple times of visual data,which often leads to large time consumption of the algorithm,and it is difficult to ensure the real-time performance of the system.In this paper,we propose a real-time visual-inertial SLAM system based on multiple cameras.The system focuses on solving the problem of time-consuming multi-camera SLAM system,and has high real-time performance and robustness under different lighting conditions and occlusion conditions indoors and outdoors.The main research contents of this paper are as follows:(1)Modeling of multi-camera system.Aiming at the problem that mainstream visual SLAM methods do not support multi-camera systems,this paper uses the rigid-coupled multi-camera system model to establish the observation model of multiple cameras.By introducing a virtual optical center,the model sets multiple cameras without common view area into a virtual camera.Using the external parameters from the original single camera to the virtual camera,the visual observation of a single camera is transformed into the virtual camera to fuse the observation information of multiple cameras.(2)Research on real-time multi-camera front-end algorithms.Aiming at the problem of time-consuming in multi-camera vision front-end,this paper proposes a feature point tracking algorithm based on hardware optical flow field.The algorithm mainly uses the optical flow field generated by the frame matching information to track the feature points,so as to reduce the time consumption of the algorithm.At the same time,in order to further improve the real-time performance,this paper uses multithreading technology to realize the parallel processing of the vision frontend algorithm.Experimental results show that the hardware optical flow feature tracking algorithm takes less time than the normal optical flow tracking algorithm,and the multi-threading technology will improve the real-time performance of the overall front-end algorithm.(3)Aiming at the problem of large amount of calculation of sliding window optimization in visual SLAM system,an efficient solver algorithm for sliding window optimization is proposed.Current sliding window optimizations of open source SLAM systems are all computed using the general solver algorithm.These algorithms are oriented to general,nonlinear optimization problems.When they are directly used to solve sliding window optimization problems,there are often a lot of redundant calculations.Based on the in-depth analysis of the steps of sliding window optimization algorithm,this paper effectively reduces the amount of calculation of sliding window optimization through the optimization solution of Schur complement matrix and the preprocessing of redundancy calculation,so as to improve the real-time performance of the system.(4)An efficient real-time multi-camera SLAM sliding window solver is designed and implemented based on the LM optimization algorithm,including the algorithm steps,software structure,internal and external interface functions of the solver.The complete code of the designed efficient solver is written in C++ programming language.It is also applied to the proposed multi-camera SLAM system.(5)System experiment and analysis.The performance experiment of the complete system constructed by the research content of this paper is carried out,and the results are analyzed.