Application Research Of VI-SLAM System Based On Fish-eye Camera Model

With the rapid development of information technology and artificial intelligence,unmanned systems and mobile robots begin to enter people's daily life,and play an important role in mixed reality,emergency rescue and unmanned mission,the key technology,Simultaneous Localization And Mapping(SLAM),has attracted much attention of researchers.With the advantages of low cost and easy deployment,visual SLAM system is more likely to be used in daily life.However,there are still some problems in some application s cenarios,such as:The original effective view angle of the camera is reduced and the overlap area between two images is narrowed by using SLAM system based on pinhole camera model and de-distortion processing,when the angle of view of the system changes too much in the environment,the robustness of the system will become worse,and even the system will fail,the system can not detect enough effective visual features,and can not match the features and restore the pose between the two frames,so that the system can not work.This paper focuses on two problems: small visual angle caused by de-distortion and insufficient visual information.the main research contents and contributions of this paper are as follows:For the first problem,we use the fish-eye camera model to replace the pinhole camera model,and then deduce the fish-eye camera model in the SLAM system,which can keep the camera's original angle of view,and then we use the fish-eye camera model to solve the Jacobian Matrix and determinant problem,it is integrated into the graph optimization framework of front and back end nonlinear optimization.In addition,considering the scale problem of the Vision System,this paper adopts the binocular camera to ensure the scale determinacy of the system,and then designs the algorithm flow of the binocular non-distortion feature matching system based on the monocular system,in order to improve the robustness of the scene,the visual system is provided with the invariability of certain scale and effective view angle.For the second problem,this paper uses a multi-sensor tightly coupled approach based on the Inertial Measurement Unit IMU(Intrins ic Measurement Unit)sensor.The IMU can be used to estimate the attitude characteristics,and the pre-integration mechanism is used to synchronize the IMU and the camera frequency.A unified objective function including monocular,binocular and inertial measurements is constructed at the back end.To provide reliable IMU position and pose information to assist the system when the vision information is weak,so as to ensure the robustness of the system in this scenario.In order to deal with these two problems at the same time,a new VI-IMU system which combines the fish-eye camera model,binocular camera and SLAM sensor is proposed,finally.Moreover,a set of hardware system is built to verify the feasibility of the fusion system research project...