Any method for automatic background-foreground-separation requires a robust model of the video background. Subspace estimation methods, such as algorithms derived from Principal Component Analysis (PCA) can be applied to a wide range of data [6] and have been reported to be especially successful at modeling lighting conditions and repetitive motion in the background [13]. On the downside they are reported to be very costly compared to competing methods. Many use cases for background subtraction methods, such as surveillance applications require algorithms that are realtime capable and can be implemented on concentrated hardware such as an off-the-shelf graphics processing unit (GPU) in a desktop PC. Thus, a way needs to be found to implement such methods in a time and memory efficient way.