Research On Simultaneous Localization And Mapping Technology Based On Multi-sensor Information Fusion

Simultaneous localization can recover its own pose estimation through real-time sensor sampling data.Because the observation constraints of a single sensor are less,it cannot meet complex environments.For further it improve simultaneous localization of mobile devices in complex environments,a solution that use multiple sensors fusion is popular.This paper combines two sensors of camera and inertial measurement unit.According to sensor measurement principle,research a visual inertial odometer with multi-sensor fusion localization.This paper analyze the principles of the image sensor and the inertial measurement unit,considering its contain white noise during observation,the parameters of camera is calibrated,considering its drift properties of inherent cumulative,the parameters of inertial measurement unit is calibrated.The pre-integration theory of the inertial measurement unit is introduced,the model of the multi-sensor odometer is deduced,and the parameters of multi-sensor odometer is calibrated.The motion model of the multi-sensor odometer is analyzed,and the Lie group disturbance model is introduced to determine the state variables of the system.Combined with the model of the multi-sensor odometer,an optimization based on the least square method is constructed.the image sensor is constructed as an observation constraint residual based on the pixel plane,the inertial measurement unit is constructed as an observation constraint residual based on the motion model,and the motion between the sensors is constructed as A motion-constrained residual is used to push out the mathematical model of visual inertial odometer.The experimental results show the feasibility of the VIO in this paper.With the increase of time,the state variables increased linearly,and the computational complexity increased exponentially.The strategy of key frames is used to reduce the computational complexity,Combined with the model of VIO and the factor graphs is adopted to construct a global optimization model.The back-end optimization model is defined in the entire motion space.Each state variable is optimized to ensure the adaptability of estimation.The results show that the VIO in this paper is slightly insufficient in the absolute pose error.the local accuracy of the VIO is higher than that of VINS-Mono and Open VINS in the relative pose error.