Performance of a Bio-Inspired Model for the Robust Detection of Moving Targets in High Dynamic Range Natural Scenes

Flies have the capability to detect and track small moving objects, often against cluttered moving backgrounds. From both a physiological and engineering perspective, understanding this computational process is an intriguing challenge. We have developed a target detection model inspired from electrophysiological recordings of 'small target motion detector' neurons within the insect brain. Our numerical modeling represents the neural processing along a proposed pathway to this target-detecting neuron. We use high dynamic range, natural images, to represent 'real-world' luminance values that serve as inputs to a biomimetic representation of photoreceptor processing. Adaptive spatiotemporal high-pass filtering (1st-order interneurons) then shape the transient 'edge-like' responses, useful for feature discrimination. Nonlinear facilitation of independent 'on' and 'off' polarity channels (the rectifying, transient cells) allows for target discrimination from background, without the need for relative motion cues. We show that this form of feature discrimination works with targets embedded in a set of natural panoramic scenes that are animated to simulate rotation of the viewing platform. The model produces robust target discrimination across a biologically plausible range of target sizes and a range of velocities. Finally, the output of the model for small target motion detection is highly correlated to the velocity of the stimulus but not other background statistics, such as local brightness or contrast, which normally influence target detection tasks.