Pixel-wise motion detection in persistent aerial video surveillance

We present a pixel-wise metric for change detection in stabilized, persistent aerial video. The method first decomposes a raw windowed temporal signal and performs time-frequency analysis to calculate a scalar value for each pixel. From this value, the algorithm indicates whether the pixel underwent independent motion in the time window or parallax. The method outlined does not make any assumptions on the sparsity of structural motion, amount of independent movement, nor calculate any camera geometry. The algorithm is naturally parallelizable and can be applied prior to Structure From Motion (SFM) frameworks to disambiguate independent and structural motion or create a mask for object-oriented video compression.