Project Proposal: Virtual Panning

The video recording of lectures has become increasingly popular with the advent of open courseware. Traditionally this has involved the use of a cameraman to film each lecture [Nagai (2009)]. The field of computer vision provides insight into automating this process, essentially removing the recurring costs of hiring someone to film each lecture [Wallick et al. (2004)]. Automation involves detecting features in the video and taking appropriate action. Either the detection is real-time, or it is done in post-processing. While some work has been done in the area of generating lecture videos automatically, these systems have various methods of solving the problem. In general, they have in common the following related components: 1. Camera setup and inputs 2. Feature detection and object tracking 3. Event detection and rules determining appropriate camera action Our task is to create such a system. In this proposal, we will explain the reasons leading to this project, how we propose our system will work, and how we intend spending our time developing it.