Action spotting exploiting the frequency domain

In this paper we present a novel method for action spotting in a video sequence, that employs the 3D Fourier transform in order to define a 3D gradient correlation function operating at the frequency domain. In that way we achieve invariance to spatiotemporal variations and also to frame reordering. One of the most attractive features of the proposed scheme is it's high degree of computational efficiency and the fact that it can be implemented by fast transformation algorithms in the frequency domain. Experiments conducted to a publicly available database demonstrated the superiority of the proposed method in terms of more accurate estimations and significantly lower computational complexity.