Research On Vehicle Heading And Attitude Determination Using Multi-source Information Fusion

With the development of society economy, the highway traffic tends to drive highspeed and flow intensively, and the vehicle safety issue is focused worldwide. Thevehicle stability control system can effectively avoid accidents caused by vehiclerollover, sideslip and side collision, which actively intervenes and controls the engineand the brake system. However, obtaining the vehicle attitude angle is a key issue todetermine the accuracy and stability of the control system. Due to the environmentnoise and the measurement error, it can not obtain the optimal attitude angle by usingonly a sensor. In allusion to the problems mentioned above, the paper uses themulti-sensor information fusion method to provide the accurate vehicle attitudes. Thekey techniques such as the dynamic attitude solution, the error modeling of sensor andthe information fusion method are discussed in depth. Within the framework of theproposed particle filter, the output information from the gyro, the accelerometer and themagnetometer are fusioned under the unified reference coordinate system. The paperfocuses on the sample dilution and the degeneracy problem of particle filter andproposes the optimized method on the sampling, the proposal distribution and theresampling to improve the estimated accuracy of the state and the vehicle attitude. Maincontributions of the paper are shown as follows:1. The MIWOSIR is suggested in this paper. Firstly, the MIWO is presented that isbeneficial to get the optimal value in the global scope by adding an envelope of thestandard deviation for improving the convergence rate of the IWO. Then the MIWOSIRincorporates the MIWO into sampling process of the particle filter, enables the particlesto reproduce dynamically by their own fitness in nearby space and optimize the particlepopulation of optimal weight. Though MIWOSIR, particles are moved towards regionswhere they had larger value of posterior density. The Markov chain is applied forproving the global convergence of the MIWO algorithm and4benchmark functions areused to verify the strong global search ability. Simulation results demonstrate thatMIWOSIR has higher estimation accuracy and operational efficiency than otheroptimized particle filters.2. The Fourier-Hermite Rauch-Tung-Striebel based particle filter is proposed. Inthe algorithm, a bank of Fourier-Hermite RTS smoother is employed for generating theimportance density function to impel the probabilities approximating to the true state forobtaining prediction samples with higher precision. Because all observation informationis introduced into the state transition function, consequently the suggested proposaldistribution has much more overlap regions with the real posterior distribution. Besidesthat, the stochastic perturbation re-sampling is introduced that can ameliorate the particle diversity after re-sampling and reduce the computation in a certain extent. Thenew algorithm is tested on three classic non-linear and non-Gaussian models withpromising results compared with existing ones.3. The particle filter using genetic algorithm and invasive weed optimization ispresented. In the re-sampling process, the degeneracy problem is relieved by applying aselection operator of genetic algorithm to choose the optimal particles iteratively, andthen the crossover and mutation operation are implemented for the particles which arenot selected so that the diversity of particles is maintained. Meanwhile, IWO makes thenewest observations into sampling process, and optimizes the particle population withthe optimal weight. Simulation results demonstrate that the proposed method has higherestimation accuracy.4. The paper proposes the heading and attitude determination of the vehicle usingmulti-source information fusion method. It uses MARG sensor to measure the angularrate, acceleration and geomagnetic information of the vehicle, and regardes the localCartesian coordinate system as the reference coordinate system. The state error model isbuilt with the multiplicative MRPs and the angular rate error vector, the observationerror model is built with the the acceleration vector and geomagnetic vector, and anoptimal estimation fusion structure with feedback by using the improved invasive weedoptimized particle filter is built. The system takes the continuous measurement of thegyro as the attitude reference information of vehicle in short-term, simultaneouslymakes the measurement of the accelerometer and the magnetometer be the attitudecorrection information of the gyro in long term. The improved invasive weed optimizedparticle filter is used for correcting the attitude error and the gyro drift to establish anaccurate model of the attitude determination estimator, which achieve an optimalassociation of the nonlinear attitude observation. By comparing with the traditionalattitude estimation methods in experiment, the proposed method improves the vehicleattitude estimation accuracy.The numerical simulations and the experiments for vehicle heading and attitudeestimation show that the optimized strategies of the paritlce filter in this papereffectively eliminate the particle degeneration and particle impoverishment, whichimproves the estimation accuracy of the vehicle heading and attitude. It has animportant reference for application and popularization of the vehicle heading andattitude determination system using multi-source information fusion.