Building a video database of animal and background movements to investigate naturalistic camouflage breaking by animal motion

When animals move, predator vision is well adapted to exploit that motion to break their camouflage. There are suggested strategies where motion can be used as camouflage, including: (1) Motion Signal Minimization - animals adapt their motion to resemble their surroundings; (2) Protean motion - unpredictable motion hinders a predator’s estimation of prey location; (3) Punctuated motion - when prey stop and start motion unpredictably. However, few accounts note that prey are not the lone occupants of natural visual scenes and thus predators must distinguish prey motion from environment motion. It is not known which patterns of motion best enhance camouflage in a dynamic environment, nor whether prey speed or direction best enhance the camouflage techniques described above. Here we present a framework to address these avenues of investigation. We collected a natural video database of prey animals moving in natural moving backgrounds. Alongside this, we built a 3D modelled video database of forest scenes and prey movements which simulate a range of realistic predator prey interactions. We used a variety of naturalistic environments with variable landscapes, vegetation, composition, environmental lighting, wind speeds, materials, scale and viewing distances. A small prey target was designed to follow four motion patterns: straight, punctuated, sinusoidal and protean zigzag, moving at slow, medium and fast speeds (1.6mph, 3mph and 5mph). These were tested across three 10mx10m environments (default plain landscape, static forest and simulated windblown forest with 3mph wind) at a viewing distance of 5.5m. We extracted motion and found prey motion was contaminated by that of the background in windy conditions. Creating this benchmark dataset will enable a more complete testing of the effects of prey motion on motion processing mechanisms in situations more closely approximating the world. Our work presents, for the first time, a way to quantitatively analyse strategies for camouflage-from-motion.