A novel camera motion classifier for 2D-to-3D video conversion

  • R. He

Student thesis: Master

Abstract

Converting existing 2D videos into 3D videos is an important technology for televisions and broadcasting. Motion based cue is an important cue in the conversion. As different camera motions require different strategies, we propose a classifier to distinguish between camera translation and camera rotation. The classifier is realized by calculating the fundamental matrix for the frame pairs in the video sequence. The fundamental matrix is calculated from matched feature points. Therefore we examine two feature detection and matching algorithms, and select the one that achieves better classification performance. The Singular Value Decomposition for the fundamental matrix is chosen as the criteria for the classification. Further a median filter is applied to increase the accuracy of the classifier and experiments show the classifier's high accuracy in classification (100% for most test videos).
Date of Award31 Aug 2013
Original languageEnglish
SupervisorL.P.J. Vosters (Supervisor 1)

Cite this

'