Karl Haflinger1, Eric Torgenson2
1Sea State Inc., 2Finsight LLC
Electronic monitoring of full-retention trawl fisheries using video surveillance has been successfully demonstrated in past years. However, obstacles remain that could prevent widespread adoption. We are developing, and will demonstrate key aspects of, a computer system designed to address some of these issues.
Because many current EM systems require human review of large volumes of video data from multiple cameras (often three or more), labor costs associated with reviewing data for existing EM systems prevent them from being sufficiently cost competitive with shipboard observers. We addressed this problem by developing video content analysis techniques that can be used to pre-process EM video before review and identify periods for which review is unneeded or can be cursory, as well as identify key events such as gear deployment and retrieval. This approach preserves all of the video from a trip, but makes it possible to make a determination that no discard events took place after only reviewing a small part of it. By reducing the number of hours required to review a trip, we not only reduce the cost of review, but can improve the timeliness of review, and potentially reduce the amount of video data that needs to be archived. Because all of the footage from each trip is preserved, additional random audits against logbook data or other second level reviews are easily accommodated.
Many current EM systems use low resolution cameras and low frame rates. We will demonstrate the tradeoffs between high resolution/high frame rate video streams versus video streams at various lower resolutions and frame rates, and discuss the technical challenges presented at both ends of that spectrum.
