Real-time motion tracker for a robotic vision system

Abstract

Estimation of distance to nearby landscape features is an essential navigation capability for an autonomous mobile robot. The ability to track targets in video sequences, combined with the concept of motion vision, allows a robot to rely on vision-based navigation techniques and avoid using active sensors or sophisticated stereo imagers for distance measurement. We present recent developments of our work on a real-time vision system capable of estimating distance to targets by tracking their apparent motion. The algorithm is designed for implementation on chip-level parallel computer architecture and to work synchronously with an imager at very high frame rates. Test results demonstrate sufficient stability of tracking and low sensitivity to noise and interference.