Studying fish near ocean energy devices using underwater video

The effects of marine and instream energy devices on fish populations are not well-understood, and studying the behavior of fish around these devices is challenging. To address this problem, we have evaluated algorithms to automatically detect fish in underwater video and propose a semi-automated method for ocean and river energy device ecological monitoring. The key contributions of this work are the demonstration of a background subtraction algorithm that detected 87% of human-identified fish events and is suitable for use in a real-time system to reduce data volume, and the demonstration of a statistical model to classify detections as fish or not fish that achieved a correct classification rate of 85% overall and 92% for detections larger than 5 pixels. This automated processing would significantly reduce labor time and costs, compared to current monitoring methods. Specific recommendations for underwater video acquisition to better facilitate automated processing are given. The proposed automated processing and recommendations will help energy developers put effective monitoring systems in place, and could lead to a standard approach that advances the scientific understanding of the ecological impacts of ocean and river energy devices.