Background Estimation by Weightless Neural Networks

BEWiS is a background modeling approach for videos based on a weightless neural system, namely WiSARDrp, with the aim of exploiting its features of being highly adaptive and noise-tolerance at runtime. In BEWiS, the changing pixel colors in a video are processed by an incremental learning neural network with a limited-in-time memory-retention mechanism that allow the proposed system to absorb small variations of the learned model (background) in the steady state of operation as well as to fastly adapt to background changes during the video timeline.