Fast conditioning algorithm for significant zero curvature detection

Zero curvature points are commonly used as features in machine vision. Traditional approaches to zero curvature detection rely heavily on discrete curvature estimation done in scale-space, which is costly to compute. The authors report their work on achieving a quick approximation by using conditioning. The algorithm is efficient and the zero curvature points detected are stable across scales. Usually these detected locations of zero curvatures are used for initialising the coarse-to-fine matching process for object recognition. Hence, the tradeoff between their accuracy and runtime efficiency must be balanced