Spatio-temporal optimization method for determining motion vector fields under non-stationary illumination

摘要

The present paper proposes two generalized gradient-based methods. The methods assume that motion vector fields are stationary within a large temporal domain. In addition, they assume that an unknown parameter representing the effect of non-stationary illumination is changing within the same temporal domain. For approximation of the unknown parameter, one of the methods utilizes a polynomial function of time; the other assumes short time constancy of the unknown parameter within a small temporal domain. Through analysis of synthetic and real image sequences, the latter method is confirmed to be more accurate and robust.