The neuronal pathway for biological motion vision is complex and non-linear. Despite considerable research effort it has defied accurate modelling for over 50 years. Recently we proposed a computational model for the calculation of egomotion that explained a number of outstanding issues, such as reliable coding in different environments and responses to artificially contrast rescaled images. Here we varied the amount of noise to determine the robustness of the model under conditions more typical of real-world scenes. High-dynamic range panoramic images taken from various environments were used as inputs to a computational motion model of biological motion vision. Gaussian white noise was added after image pre-processing but before motion detection. The addition of noise around the levels observed experimentally, in both biology and an engineered camera system, resulted in a surprising 50% increase in the discriminability of different velocity levels over that seen in the noise free condition. The more commonly used gradient model for motion detection produced outputs so swamped by noise they were unreliable under the same conditions. While the phenomenon of stochastic resonance has been observed previously in biological and bio-inspired systems it is most commonly found in conjunction with non-linear thresholding operations, such as spike generation. These findings are unusual as they show noise being beneficial in a model of an analogue system. They also highlight the robustness of the correlation model for biological motion detection to very large levels of noise.