A Detection, Tracking, and Classification System for Underwater Images

Traditional underwater images for ecology study often include time-lapse or brief video recordings over short periods of time. More recently, with the installation of under water cabled observatories that provide constant electrical power and data connections to the seafloor, longterm video recordings are possible for the first time. These valuable recordings are essential for underwater ecology studies, particularly abundance and distribution studies but also behavioral studies. However, the analysis of these video and time-lapse recordings often becomes quickly overwhelming. In some cases, the number of underwater animal activities can be infrequent, resulting in recordings with many hours of video with few events of interest. Sometimes the human resources do not exist to analyze the recordings, or if resources are available, the strains on human attention quickly abate the efforts. To help address this issue, over the past six years an automated detection, tracking, and classification software system called The Automated Visual Event Detection and Classification System (AVEDac) has been under development at the Monterey Bay Aquarium Research Institute (MBARI). The AVEDac system is a powerful aid that is currently being used to analyze Remotely Operated Vehicles (ROVs) and deep-water cabled observatory video recordings recorded in the Monterey Accelerated Research System (MARS) observatory. It has been recently modified to also process still images recorded from Autonomous Underwater Vehicles (AUVs) and stationary cameras on the sea floor.