HR: 0800h
AN: OS51C-0576    [Abstracts]
TI: Detecting, Tracking and Classifying Animals in Underwater Video
AU: * Edgington, D R
EM: duane@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Kerkez, I
EM: ikerkez@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Oliver, D
EM: doliver@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Cline, D E
EM: dcline@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Sherlock, R
EM: robs@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Robison, B
EM: robr@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Kuhnz, L
EM: linda@mbari.org
AF: Monterey Bay Aquarium Research Institute (MBARI), 7700 Sandholdt Road, Moss Landing, CA 95039 United States
AU: Ranzato, M
EM: ranzato@cs.nyu.edu
AF: New York University (NYU), Computer Science Department, New York, NY 10012 United States
AU: Perona, P
EM: perona@caltech.edu
AF: Calfornia Institute of Technology (Caltech), 1200 East California Boulevard, Pasadena, CA 91125 United States
AB: The Monterey Bay Aquarium Research Institute (MBARI) deploys remotely operated vehicles (ROVs) equipped with high resolution video equipment. This technology enables quantitative video transects (QVTs) to be obtained providing data at the scale of the individual organisms and their natural aggregation patterns. QVTs are a sophisticated means of sampling that has recently replaced conventional methodologies and significantly advanced studies in animal diversity, distribution and abundance. The method currently used to analyze QVTs, however, is labor intensive and costly, reducing the amount of data analyzed from the ROV dive and thus limiting marine ecological research. An automated program for detecting and classifying organisms in the video would address these concerns. Video frames are processed with a neuromorphic-selective attention algorithm, modeled after the human vision system. The candidate locations identified by this module are subject to a number of parameters that when combined with successful tracking determine whether detected events are deemed "interesting" or "boring". "Interesting" events are marked in the video frames. The interesting events undergo further processing with a Bayesian classifier utilizing a Gaussian mixture model to determine the abundance and distribution of a representative benthic species. Presented data details the comparison between automated detection of organisms and program classification of Rathbunaster californicus in video footage with professional annotations.
UR: http://www.mbari.org/aved
DE: 0540 Image processing
DE: 4815 Ecosystems, structure, dynamics, and modeling (0439)
DE: 4894 Instruments, sensors, and techniques
SC: Ocean Sciences [OS]
MN: Fall Meeting 2005