Frequency-domain based video irregularity analysis

A typical task of camera systems in industrial environments is to observe cyclically recurring production patterns. This paper deals with the comparison of videos with the aim of revealing irregularities from these repeating patterns. We suggest interpreting each pixel of a video as a temporally varying signal. We show that irregularities can be detected easily in the frequency domain. Compared to existing methods, e.g. based on structure or color, we obtain a nearly lossless comparison that is able to deal with the stroboscope effect. The result is a 2D-irregularity map that shows deviations from the regular production process on a pixel-by-pixel basis.