Variational Bayes Kalman Filter for Joint Vehicle Localization and Road Mapping Using Onboard Sensors

This paper addresses the joint vehicle-state and road-map estimation based on global navigation satellite system (GNSS), camera, steering-wheel sensing, wheel-speed sensors, and a prior map. Because prior maps, e.g., generated from mobile mapping systems, are updated infrequently and do not capture high-frequency events such as road construction, we include the map parameters in the estimation problem. Both GNSS and camera measurements, such as lane-mark measurements, have noise characteristics that vary in time. To adapt to the changing noise levels and hence improve positioning performance, we combine the sensor information in a noise-adaptive variational Bayes Kalman filter to jointly estimate the vehicle state, the parameter vector of the map, and the measurement noise. Simulation results indicate that the method can accurately adjust the measurement noise to the environmental conditions and thereby correct for errors in the prior map while providing accurate vehicle positioning.