Automatic capture and creation of presentations from multimedia lectures
This dissertation describes the design, implementation, and testing of Presentations Automatically Organized from Lectures (PAOL), a content capture system that automatically captures lectures. The captured lectures are organized into a presentation that includes a video, a thumbnail-based table of contents, and windows containing the material currently being presented on computer and whiteboard. These presentations are searchable using the table of contents and give the viewer access to all material presented in the lecture. The capture process occurs transparently to the lecturer with the only requirement being that the lecturer wear a wireless microphone. No training or installation of special software is required. These presentations are created to give students access to class content after the class and through the table of contents allow students to review specific portions of the material presented.
PAOL is divided into 3 subsystems: video creation, computer capture, and whiteboard capture. Details are given on the design and implementation of each subsystem and the success of different approaches to subsystem tasks. All subsystems rely on image processing techniques to create the content of the output presentation.
Image processing techniques are used to create a video of the lecturer from fixed-view high-resolution cameras focused on the whiteboard. Image processing techniques are also used to identify significant changes on the computer and whiteboard in order to form the table of contents. The whiteboard images are also processed to remove the instructor, heighten contrast, and improve clarity.