Transforming Camera Geometry to A Virtual Downward-Looking Camera: Robust Ego-Motion Estimation and Ground-Layer Detection

This paper presents a robust method to solve the two
coupled problems: ground layer detection and vehicle egomotion
estimation, which appear in visual navigation. We
virtually rotate the camera to the downward-looking pose
in order to exploit the fact that the vehicle motion is roughly
constrained to be planar motion on the ground. This camera
geometry transformation, together with planar motion constraint,
will: 1) eliminate the ambiguity between rotational
and translational ego-motion parameters, and 2) improve
the Hessian matrix condition in the direct motion estimation
process. The virtual downward-looking camera enables us
to estimate the planar ego-motions even for small image
patches. Such local measurements are then combined together,
by a robust weighting scheme based on both ground
plane geometry and motion compensated intensity residuals,
for a global ego-motion estimation and ground plane
detection. We demonstrate the effectiveness of our method
by experiments on both synthetic and real data.