Robust scene reconstruction from lines and points in three uncalibrated color images

A new robust scene-reconstruction algorithm using points and lines in three uncalibrated color images is introduced. The algorithm is robust because it discards outliers at the same time as it estimates the geometric constraints. The algorithm consists of three components: a point-tracking algorithm for color images; a line-matching algorithm for color images and a robust algorithm for computing the trifocal tensor (also called the trilinear tensor). These three components are carried out iteratively within the scene-reconstruction algorithm until no significant improvement can be achieved. Cameras are assumed uncalibrated and camera poses are assumed unknown.