The document proposes a perceptual mixture-of-Gaussians approach to background subtraction that adapts dynamically based on principles of human visual perception. It models the relationship between detectable foreground pixels and background pixels based on Weber's law of just-noticeable differences. This results in piecewise linear relationships and different adaptation speeds for scotopic and photopic vision. An evaluation of 50 test sequences shows the proposed approach achieves more realistic background value prediction, higher stability across scenarios, and superior detection quality compared to other statistical methods.