3D Modeling from Image Sequence: Lens Distortions and Partial Surface

S.S. Wong and K.L. Chan (PRC)


camera calibration, lens distortion, surface reconstruction


This project aims to reconstruct a high quality 3D object model from an image sequence captured by a mobile camera. The whole process consists of camera calibration, volumetric model reconstruction, surface model reconstruction and texture mapping. Only the first two steps are reported in this paper. Our research has two major contributions. Firstly, we have designed a new calibration pattern and camera calibration method that can acquire the object view and estimate camera parameters simultaneously with on-line lens distortions compensation. Secondly, we have developed a new silhouette-based volumetric model reconstruction method such that a better partial surface can be generated for each view. The camera can be calibrated either by Tsai's coplanar method or non-coplanar method. With both radial and tangential lens distortions compensated, the normalized calibration error is reduced by 40% as compared to the calibration with only radial lens distortion compensation. The partial surface generation algorithm, by checking for color consistency among the neighboring views, can accurately reconstruct object shape with concavity.

Important Links:

Go Back