Recovering Metric Information from Old Monocular Video Sequences

In this paper we analysis monocular sport videos, digitized from old VHS tape, where players and camera are moving. It is very difficult to recover 3D information of the scene from these sequences, due to noisy measurements, low quality of the images, very small translation of the camera and almost no information concerning its parameters. Classic stereo algorithms are not always suitable and perspective models can easily fail, due to zooming effects and baseline absence. We calibrated a monocular sequence with perspective and projective approaches, comparing the methodologies and the results. Moreover, applying some geometric image invariants, metric measurements of the movements of the players are also recovered.