| High precision real-time vehicle navigation and positioning is the core technology of vehicle networking research,such as driving assistance and other vehicle driving safety control,traffic guidance management,vehicle road collaborative control,and other intelligent transportation.In recent years,with the rapid development of intelligent transportation,vehicle navigation system has been holding a huge market demand worldwide.At the same time,the development of sensor information fusion technology also makes it possible to realize the integrated system of autonomous navigation in vehicles.At present,the modern multi-sensor integrated systems applied in most civilian navigation systems are usually composed of GPS receivers,IMU,and/or other sensors(such as odometers,cameras,etc.).However,the driving environment of vehicle is complex and changeable,which has high requirements for the accuracy and reliability of the navigation and positioning system,and the expensive navigation and positioning system is not conducive to large-scale applications.With the continuous improvement of low-cost sensor technology,more and more research and development efforts aim to establish a cost-effective navigation system without sacrificing navigation performance.Therefore,the development of vehicle navigation and positioning system based on low-cost sensors is of great significance.In this dissertation,a Generic Multi-sensor Integration Strategy(GMIS)based on low-cost IMU is proposed to meet the high-precision positioning requirements and price limitations of vehicle navigation and positioning system.The multi-sensor integrated system is mainly designed and implemented by combining multiple low-cost IMUs with GPS receivers.It has established a good model for the applications of low-cost and high-performance multi-sensor inertial navigation systems and laid a foundation for the following academic research work.The main features of GMIS are as follows: 1)The core system model is established based on rigid body kinematics.2)In Kalman filtering,the output measurements of all sensors(including IMU)are directly taken as the raw measured values without differential operation.Compared with the traditional Error-state based Integration Strategy,the main competitive advantages of GMIS can be summarized as follows: 1)Due to the increase of redundancy in measuring angular velocity and acceleration of the rigid body,the influence of IMU measurement noise on the final navigation solution can be effectively alleviated.2)Under GMIS,the state vector and measurement vector of Kalman filter can be extended to integrate the measured values of multiple inertial sensors and all other types of sensors.3)The raw sensor data(measurement noise analysis)and virtual zero-mean process noise(process noise analysis)can be directly analyzed through the corresponding measurement residuals of single measurement value and process noise.This dissertation will mainly focus on the implementation and improvement of GMIS and the improvement of Kalman filter stochastic model to carry out the research work.The main research work of this dissertation is as follows:1.In the traditional integration strategy,the inertial sensor adopts a priori error model,while the time-varying noise model of low-cost IMU has high temperature sensitivity and dynamic excitation,which is extremely unfavorable to the integrated system.Different from the traditional integration strategy,this dissertation has made full use of the kinematics principle of the rigid body and established a real three-dimensional motion trajectory model as the core system model of Kalman filter,which can realize strict trajectory tracking within IMU data interval or specific finite time interval and fundamentally alleviate the impact of low-cost IMU time-varying error on inertial navigation solution.At the same time,in GMIS the measurement values of all sensors(including IMU)can directly and independently participate in the measurement update of Kalman filter.Therefore,this dissertation has modeled the measurements and system errors of all IMU arrays in the Kalman filter independently,so that the heavy dependence of the inertial navigation mechanization on the IMU measurements in the conventional integration strategy is released.Through the analysis of GMIS performance boundary,the feasibility and reliability of GMIS also are verified.2.Since the vehicle will inevitably experience different forms of motion in the process of movement and the maneuvering process is controlled by humans with little a priori information,it is difficult to accurately describe the motion process with mathematical formulas.The general three-dimensional trajectory model cannot realize the smooth transition between different motion forms of vehicles.This will seriously affect the accuracy of the navigation and positioning solution.Aiming at the smooth transition problem,this dissertation has proposed two schemes to achieve smooth transition between different motion states.The first solution is to make full use of the changes in the linear and circumferential velocity,acceleration,and the system attitude to develop a practical mechanism to switch between different motion forms.The second solution is to establish a three-dimensional motion trajectory model based on the “Current” Statistical Singer Acceleration Model(CSSAM)to achieve smooth transition of vehicles.Through the experimental simulation data,the effectiveness of the two proposed schemes is verified.3.There exist various error sources in Kalman filter.To analyze these error sources independently,this dissertation has adopted a new method to study the random information in Kalman filtering.Namely,the system innovation vector is divided into three groups of independent residuals: the residuals of the measurement vector,the residuals of the process noise vector,and the residuals of the predicted state vector exclusive of the effect of the process noise.Thereby a new model of linear optimal filtering is established,and also the related Kalman filter quality analysis methods are introduced.4.It has always been a huge challenge to improve the a priori stochastic model of the process noise vector and measurement noise vector in Kalman filter.To eliminate the impacts of the random error in the sensor measurement and process noise in inertial navigation,this dissertation has proposed a generic random model adjustment method,which takes the advantages of the measurement residuals,process noise residuals,and measurement redundancy contribution at all times.Moreover,this method has improved the existing posteriori Variance-covariance Component Estimation(VCE)algorithm,therefore realized the estimation of the variance components related to the process noise and measurement noise vector.At the same time,the improved VCE algorithm is used to calculate the measurement weight of each sensor(especially IMU).Thereby the ratio of each measurement value is determined in the fusion algorithm and each measurement value is allocated reasonably,which can simplify the measurement model,reduce the amount of calculation,and greatly improve the performance of GMIS.5.Finally,this dissertation has designed and conducted effective vehicle road tests,collected experimental data,and verified the effectiveness of the key technologies proposed in this dissertation.The focus of the verification has been laid on the significant performance improvement of the generic multi-sensor integration strategy by applying the improved posteriori variance component estimation algorithm.The test results has fully proved the reliability and feasibility of the low-cost IMU-based generic multi-sensor integration strategy proposed in this dissertation. |
PDF Full Text Request