Video tomography: an efficient method for camerawork extraction and motion analysis

This paper proposes a new, efficient and practical way to extract lens zoom, camera pan and camera tilt information using modified motion analysis. The proposed method is called the Video Tomography Method (VTM), in which tomographic techniques are introduced into a motion estimation algorithm. By using the VTM, one is able to visualize motion as a spatiotemporal flow for motion analysis. The VTM is an extremely robust [resistant to noise] method for estimating camera operation due to its tomographic nature. The practicality of this type of motion estimation and analysis is confirmed by the results of our simulations and experiments in testing the prototype platform with a low quality video source. Other possible applications that might use extracted motion data are discussed. This method is targeted towards video handling applications that attribute extracted motion data into a video index. It will enhance the process of editing and browsing structured video, and will allow the visualization of scenes spatiotemporally so that video may be accessed intuitively and spatially in relation to its temporal location. This type of access is a new interface for structured video. Scene reconstruction techniques can be extended to apply to the problem of reconstructing occluded images and resolution enhancement.