Motion stereo for mobile robots

Abstract

A method allowing a mobile robot to measure accurately its distance from external objects is introduced. It only requires a single uncalibrated camera and a dynamic vision system. The method was evaluated in an optical laboratory and also in outdoor-experiments. Even in outdoor-experiments with a vehicle moving at a speed of 30 km/h, errors of less than 1% of the true distance were achieved in real time. The accuracy achievable with the method depends on the nature of the image features used as a basis for the measurement. Laboratory experiments have been performed to investigate quantitatively the characteristics of the feature detectors under conditions of reduced lighting and, consequentially, reduced SNR, and with image features not matching the feature detectors perfectly. The best results can be achieved by using edge-like features. Point-like features do have certain desirable characteristics, but point detectors are more sensitive to noise than edge detectors. Hence, the accuracy achieved with corner detectors was less than the one achieved with edge detectors. Nevertheless, even with corner detectors the resulting error was less than 1%, despite the fact that the image of the target object did not contain good corners.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>