Behringer and Hoetzl 1994 : The vision system for autonomous road vehicle guidance,
which has been developed during the last years at the UniBwM, is currently
being ported from 5 ton vans to a family sedan (Mercedes 500 SEL). The cruising
speed of 120 km/h, which the system is designed for, requires a further maximal
look-ahead distance for road and obstacle recognition. For achieving high
robustness, it is necessary to estimate the camera pitch angle as an additional
state parameter. In the video image, the mapped lane width is a useful cue for
pitch angle estimation, when the real lane width is well known. Therefore, it
is necessary to estimate the real lane width in the general case of an unknown
and changing lane width.
In this paper an approach for simultaneous estimation of pitch angle and lane width from the road image is presented, both based on Kalman filtering. The pitch angle is modeled by a damped harmonic oscillator with a given frequency ω0, a time constant, and a pitch offset. For a change in the lane width, a linear dynamical model is assumed. The estimators are validated by numerical simulations. Evaluation of video scenes, shot from the sedan vehicle, show good results for pitch and lane width estimation.