Integrating Tsai's Camera Calibration Algorithm with KLT Feature Tracking

Rodrigo MinettoNeucimar J. LeiteJorge Stolfi

We describe a robust and accurate algorithm to track selected features of a 3D scene in a video recording. The algorithm uses a synergistic combination of Roger Tsai's camera calibration (TCC) algorithm and the Kanade-Lucas Tomasi (KLT) feature finder. Instead of trying to detect and reject outliers, we modified Tsai's algorithm to use confidence weights which depend on the quality of the match obtained by the KLT finder. Conversely, we use the TCC-calibrated camera parameters to predict the position and shape of each feature. As a result, our algorithm can usually recover a feature after an occasional KLT failure, or even after it has been occluded for many frames.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: