Analysis of distributed smart camera calibration accuracy using minimal data transmission

An algorithm for accurately estimating camera overlap in a distributed smart camera network with minimal data transmission was developed and evaluated, as a preliminary step towards three-dimensional calibration. The algorithm incorporates SURF (Speeded-Up Robust Features) to find matching features in live, low-quality images provided by web cameras. To reduce transmitted data, the number of extracted features was reduced, and feature descriptors were compressed using singular value decomposition. Reliable overlap estimation accurate to within 9 pixels was achieved while sending just 30 kilobytes per captured image.