Abstract

Underwater Video Spot Detector (UVSD) is a software package designed to analyze underwater video for continuous spatial measurements (path traveled, distance to the bottom, roughness of the surface etc.) Laser beams of known geometry are often used in underwater imagery to estimate the distance to the bottom. This estimation is based on the manual detection of laser spots which is labor intensive and time consuming so usually only a few frames can be processed this way. This allows for spatial measurements on single frames (distance to the bottom, size of objects on the sea-bottom), but not for the whole video transect. We propose algorithms and a software package implementing them for the semi-automatic detection of laser spots throughout a video which can significantly increase the effectiveness of spatial measurements. The algorithm for spot detection is based on the Support Vector Machines approach to Artificial Intelligence. The user is only required to specify on certain frames the points he or she thinks are laser dots (to train an SVM model), and then this model is used by the program to detect the laser dots on the rest of the video. As a result the precise (precision is only limited by quality of the video) spatial scale is set up for every frame. This can be used to improve video mosaics of the sea-bottom. The temporal correlation between spot movements changes and their shape provides the information about sediment roughness. Simultaneous spot movements indicate changing distance to the bottom; while uncorrelated changes indicate small local bumps. UVSD can be applied to quickly identify and quantify seafloor habitat patches, help visualize habitats and benthic organisms within large-scale landscapes, and estimate transect length and area surveyed along video transects.

Department

Center for Coastal and Ocean Mapping

Publication Date

9-2005

Volume

3

Journal Title

IEEE Oceans

Pages

2189-2192

Conference Date

Sep 18 - Sep 23, 2005

Publisher Place

St. Andrews, New Brunswick, Canada

Publisher

IEEE

Digital Object Identifier (DOI)

10.1109/OCEANS.2005.1640089

Document Type

Conference Proceeding

Download

Share

COinS