Research On Vehicle Motion Estimation System Based On Improved Kalman Filter

With the rapid development of autonomous driving technology,to make the decision system of vehicle to be more intelligent,the motion estimation system need to be more and more accurate and robust.With the simple structure,low cost,and the flexibility,the loose integrated navigation system combined with Global Navigation Satellite System,Strapdown Inertial Navigation System and other sensors,can adapt to all kinds of scenarios and output the accurate motion information in high frequency.The integrated navigation system is one of the mainly research direction of the motion estimation system,and it will play a crucial role in the development of autonomous driving technology.This thesis discusses vehicle motion estimation based on Kalman filter system respectively in the inertial sensor error model and error parameters calibration method,the sensor output random error suppression method and the adaptive UKF algorithm based on dynamic sampling window width with integrity constraint conditions three aspects.The research methods and innovations are shown below.1.A low-cost error calibration method for inertial sensor based on matrix decomposition method and 19 position scheme is proposed.Firstly,the sensor error model is determined through the fitting experiment with the output of the sensor,and then the matrix decomposition decouples the error coupling term in the calibration process.Finally,the calibration of the error parameters of the sensor is completed quickly and accurately by using the 19 positions scheme.2.In view of the low cost inertial measurement unit output information of low signal noise ratio(SNR)problem,a wavelet denoising method based on improved threshold function is proposed.By adjusting the parameters,the character of the improved wavelet threshold function was not confined to a single soft threshold function or hard threshold function,make the wavelet coefficient function after screening could still completely express the characteristics of the original data to identify the useful data and noise.3.An adaptive UKF based on dynamic sampling window is proposed to solve the problem that trackless kalman filter algorithm requires high accuracy of system noise statistical characteristics.At the same time,the adaptive sampling window width is also dynamically adjusted to improve the matching of the adaptive sampling process to the real noise characteristics of the current system.In order to solve the problem of error divergence when no reference information is input into the motion estimation system,the vehicle motion characteristics are studied and analyzed,and an improved adaptive algorithm based on non-integrity constraints is proposed,so as to maintain a high level of accuracy in the whole operating range.All above methods are tested and verified in simulation.And then,these methods are tested with the data come from the vehicle motion estimation system independently made by our team.The result shows that the method proposed in this thesis is effective and these methods could provide certain reference for the vehicle motion estimation system research.