Understanding local flounder and cod populations
The School of Marine Science and Technology (SMAST) was developing a new, better way to conduct population surveys in the New England groundfish fishery. By filming fish passing through the cod end of an open net, they could avoid having to pull up a closed net to count fish by hand. Initially they were using human reviewers to watch the film and record how many fish of each species were passing through. When they started getting more data than their reviewers could handle, they hired CVision AI to count the fish automatically.
CVision AI developed a human in the loop (HITL) tracking application that allowed our algorithms to team up with reviewers to quickly create perfect tracks of the fish passing through the video frame. We then built a state-of-the-art multi-object tracking algorithm that learned from the human reviewer, resulting in a fully automated algorithm that could count fish within an error of 5%. Tracks were then classified by species using a novel sequence level classification algorithm.
