Spectral unmixing for tracking human motion in low resolution imagery
Vivekanand Govinda, Jason F. Ralph, Joseph William Spencer, John Yannis Goulermas · Journal of Physics Conference Series · 2007
This paper discusses how colour/hue information can be used to track people in image sequences received from very low resolution sensors. The main colour spaces such as Normalised RGB, HSV, CIE L*a*b*and YCbCr have different advantages and disadvantages when applied to tracking in very low resolution imagery, where the objects of interest may be sub-pixel in size. Tests have been carried out to find out the best colour space or combination of colour spaces. Spectral unmixing algorithms, commonly used in remote sensing and multi-spectral image processing, were used to assist object tracking algorithms by decomposing pixel colour information into fractional contributions from backgrounds and skin.