Model based estimation of ego-motion and road plane using vehicle-mounted camera

This paper presents a new image feature based method to estimate ego-motion and relative pose to the road plane for a road vehicle-mounted camera. We introduce a restrictive motion model exploiting several constraints arising from the construction of road vehicles, thus significantly reducing parameter space and increasing overall estimation robustness. The model enables estimation of relative camera pose by optimization of four parameters. By adding a fifth parameter additional estimation of the road plane normal in a joint optimization process is possible. The proposed algorithm has been tested in authentic traffic environments and its superiority is shown in comparison to existing methods.