Robust Text Activity Detection from Videos of Computer Monitors

Publication Year:
Usage 4
Abstract Views 4
Repository URL:
Saravanan, Krithika
artifact description
The project introduces a robust method for detecting text regions in videos of computer monitors. The basic problem comes from the need to assess middle-school student learning when using a low-cost Raspberry Pi computer. To assess learning, we need to detect the text entered by the students without using video capture software that would severely limit the performance of the Raspberry Pi. We describe a robust approach that uses projections and Quadtree Decompositions to detect the monitor, the windows boundaries, and the individual characters. We provide visual assessment on ten challenging videos that exhibit non-uniform illumination, monitor shaking, and variable viewing angles.