Vehicle speed estimation based on a modified particle filter algorithm

For conventional vehicle speed estimators designed based on the particle-filter algorithm, the estimation performance deteriorates if the proposal distribution is inconsistent with the actual distribution. In this paper, an improved particle-filter speed estimator is proposed to tackle this problem by modifying the proposal distribution. Firstly, the state transition equation and the observation equation of the system are established based on vehicle kinematics and sensor characteristics. Then, the difference between sensor measurements and particle state values is employed to design a correction term for the proposal distribution, simultaneously adapting the process noise in the state transition equation. Finally, simulation validation is conducted using CarSim-Simulink co-simulation platform under the double-lane change and the sine-wave steer input maneuvers. Compared with the adaptive particle filter, the proposed estimator shows reductions of 40.25% and 55.71% in the mean absolute deviations (MAD) of the estimated longitudinal velocity and the estimated lateral velocity, respectively, under the double-lane change maneuver; and under the sine-wave steer input maneuver, the reductions are 47.00% and 41.21%, respectively.