Research On Camera-Lidar-IMU Fusion Based SLAM

In recent years,unmanned systems have been widely used.Their tasks and environments become more and more complex,which lead to higher requirements for localization and perception.At the same time,with the development of embedded hardware and sensor technology,sensor redundancy has become an important means to improve the accuracy and reliability of perception and localization.How to give a full play to the advantages of each sensor to improve the performance of SLAM algorithm has gained large attention.Due to its simplicity,low cost and semantic information,visual SLAM has been well applied in automatic drive,drone,etc.Lidar SLAM is high reliable,high precision,and can adapt to large-scale scenes.A SLAM algorithm based on multi-camera vision,Lidar and IMU is proposed in this paper as to combine the advantages of those sensors.The main contribution of this paper are as follow:1.A method of circular vision SLAM based on multi-camera was proposed.In order to obtain a larger field of view Angle and improve the adaptability of the system to the environment,this paper adopts multiple cameras to sense visual information.In order to obtain quick initialization and larger field of view angle,the monocular camera and binocular camera has been used in this paper,so as to constitute circular vision around the carrier.In order to limit the number of features in sliding window optimization and solve the problem of feature point difference,this paper proposes an information matrix construction based on DF(Depth Filter).2.A method of fusing the Lidar data and multi-camera CVIO(Circular VisualInertial Odometry)is implemented.We use the point cloud data for further optimization of key frames,so as to improve the algorithm accuracy.A visual Bag of Word algorithm is used to detect loop candidate frame,which improves the accuracy of loop candidate result.3.Simulation datasets have been used to verify and analyze the performance of each sub-module.A real sensor platform was designed and built to verify the accuracy and robustness of the algorithm in real environment.