Development and Characterization of an Underwater Motion-Triggered Camera Trap

Underwater camera systems typically employ interval-based image capture rather than motion detection. One of the difficulties with motion detection underwater is that infrared sensing, such as is common in terrestrial cameras, is limited due the high attenuation of infrared in water. Here, an alternative approach is presented utilizing a low-power, low-resolution, microcontroller-based machine vision system to detect motion through image differencing. The implementation of the system is presented, and it is characterized in terms of power and motion detection. The image differencing threshold and sampling speed of the detection camera are the primary system variables. Through experimentation, these are shown to be highly environmentally dependent and factors such as target size, environmental contrast, ancillary motion, and target speed are important to consider. Positive capture rates as high as 80.5% with a false positive rate of 12.1% were observed when tested with a known video dataset containing fish against a hard coral background. Missed positive identification gaps were measured and, in certain environments, demonstrate that missed targets are often captured in subsequent images. Discussion of operational considerations are presented as well as future efforts including background averaging to remove small oscillatory motion and the use of classifiers to specifically detect fish as well as motion.