In the paper, we present a novel approach to generate
the rough height map with the inaccurate GPS pose. The
vehicle position is estimated by matching Sift feature
points from the prior map with the ones in the current
scans. The Kalman filter integrating measurements from
IMU ensures the smooth and stable localization. The
experiments of 3.0 km test dataset in urban environment
show our method can overcome the shortcoming of
sudden jumping of GPS position to some extent and
finally achieve real-time localization with half meter
error.
For next steps, owing to the relatively low resolution
of the map in the unstructured rural environments, we
will extract the features, for example, line or plane
descriptor with more robust discrimination to align and
localize the vehicle.