Modeling the effects of distortion, contrast, and signal-to-noise ratio on stereophotogrammetric range mapping

Abstract

Stereophotogrammetry typically employs a pair of cameras, or a single moving camera, to acquire pairs of images from different camera positions, in order to create a three dimensional ‘range map’ of the area being observed. Applications of this technique for building three-dimensional shape models include aerial surveying, remote sensing, machine vision, and robotics. Factors that would be expected to affect the quality of the range maps include the projection function (distortion) of the lenses and the contrast (modulation) and signal-to-noise ratio (SNR) of the acquired image pairs. Basic models of the precision with which the range can be measured assume a pinhole-camera model of the geometry, i.e. that the lenses provide perspective projection with zero distortion. Very-wide-angle or ‘fisheye’ lenses, however (for e.g. those used by robotic vehicles) typically exhibit projection functions that differ significantly from this assumption. To predict the stereophotogrammetric range precision for such applications, we extend the model to the case of an equidistant lens projection function suitable for a very-wide-angle lens. To predict the effects of contrast and SNR on range precision, we perform numerical simulations using stereo image pairs acquired by a stereo camera pair on NASA’s Mars rover Curiosity. Contrast is degraded and noise is added to these data in a controlled fashion and the effects on the quality of the resulting range maps are assessed.