Marker-less computer vision tracking for agmented reality

Abstract

A real time marker-less computer vision tracker designed for Augmented Reality is presented. It obtains accurate camera ositions and orientations relative to well-patterned planar surfaces in real time, without the use of artificial markers. It elies on Efficient Second-order Minimization (ESM) of the pixel intensity errors over a large image region to perform rame-to-frame tracking. A sub-grid gradient-based reference image selection process and an illumination model are roposed to improve the robustness and efficiency of ESM tracking. The proposed illumination model allows ESM racking to robustly handle shadows, glares, general ambient light changes and partial occlusion, while increasing the ccuracy and processing speed. Evaluation using video sequences of tracking of real objects is presented. A featuretracking omponent is added to automatically initialize and recover ESM tracking. © 2010 IADIS.