Creating a Virtual Cinematographer for the Automation of Lecture Recording

The increase in camera quality and reduction in price has yielded an opportunity for Universities to install cameras in lecture venues and record the lecturers’ presentations for students to review at a later stage. This paper explores the reduction of the 4K video recordings to a bandwidth conserving 720p stream using cinematographic heuristics and rules to guide the extraction of a smaller window from the larger video frames. This small window effectively creates a region which can be panned and zoomed around each high-resolution 4K frame, as if moved by a “virtual cinematographer”. The camera motion in the output video should, as closely as possible, resemble the panning that a human camera operator would produce. The results were shown to students who compared 2 video clips from the same timestamps but produced from different configurations of the Virtual Cinematographer. These students then chose which of the 2 clips they preferred and were asked to answer a survey in order to determine why they preferred the video they chose. The results are still being analysed but they show promise.