Generalizing inverse perspective

Inverse perspective mapping schemes have been presented in the past as frameworks in which autonomous navigation could be performed on flat surfaces, using optical flow as the main percept. We propose a generalization of these models for uneven terrains. Given an inclinometer indicating the direction of gravity and some simple odometry parameters, the visual and positional information are combined to reconstruct the 3D elevation map of navigational surfaces within the visual field. Camera transformations are applied directly on the image plane. We provide noise and sensitivity analysis using random, zero-mean Gaussian noise to determine the robustness of the proposed model.