Research On SINS/Visual Odometer Integrated Navigation System Based On Low Cost MEMS

The Strapdown Inertial Navigation System(SINS)is capable of outputting comprehensive navigation parameters without the aid of external references,and has a strong autonomy,but its error accumulates over time.At present,with the rapid development of military and civilian use,the demand for low-cost sensors is continuously improved.Although the MEMS-based strapdown inertial navigation system can complete the output of inertial navigation information,the error quickly diverges due to its strong device noise cannot be used independently.Its low cost and small size make it a very wide range of applications.How to improve accuracy is a scientific direction and worth studying.For some indoor application environments,or when the GPS signal is not available,a scheme combining visual odometer(VO)and inertial navigation is proposed.The visual odometer uses the camera as a sensor to capture the change of the position of the object in the shooting picture and reflect it on the image.According to the correspondence between the pixel points between adjacent frames,the camera motion can be recovered and has strong autonomy.In this paper,based on multi-rate Kalman filter,the inertial navigation information is combined with the monocular visual odometer information,and the inertial information is corrected by the position and velocity output by the odometer.The main research contents are as follows:Study the external field calibration method of MEMS inertial devices,analyze the device error,and establish calibration equations for gyroscopes and accelerometers.Derive a solution error model for low-cost MEMS SINS for large device noise.For the omission earth rotation rate and the angular velocity caused by the carrier motion,the new error equation is established.Considering the actual application scene,the location information is represented by the form of latitude and longitude conversion into a coordinate system position vector.MEMS simulation through the simulator,including data generation and solution.Analyze the imaging principle of the camera,analyze the calibration principle of the camera in consideration of the lens distortion,and perform calibration experiments on the actual camera.On this basis,the implementation methods of monocular visual odometer are studied,including feature extraction,feature tracking and motion calculation.Select the appropriate solution based on quickness and accuracy analysis of each step of its implementation.Using the FAST method,and the optical flow is used to track and match the features.To satisfy the assumption of gray uniformity and small motion,a 4-layer image pyramid is established for iterative calculation.And complete the physical experiment verification on the smart car platform,and the simulation verification based on the POV-Ray rendered image.Design the integrated navigation system with inertial/visual fusion based on the realization of the subsystem.Since the INS data update frequency is high and the vision is low,multi-rate Kalman filter is used for data fusion.The inertia/visual alignment is performed before fusion,and the inertial position integral term between several sets of images corresponds to the visual measurement value to restore the true scale of the monocular camera.The velocity and position information output by the monocular visual odometer is used for measurement,and the inertial navigation error is corrected and implemented by simulation.The combination of inertia/visuality is completely autonomous,and is of great significance for indoor mobile robots and military carriers that require strong concealment.