Marine Analytics Using Computer Vision

Purpose:

Understanding the marine environment is a key component to a more sustainable Earth. Technologies to automate data collection and analysis of the marine environment are necessary. Underwater cameras and AI (in the form of computer vision algorithms) are predicted to play major roles in this regard. This research project starts in a recently established underwater camera setup that captures video in various conditions. The project aims to create an underwater computer vision system that can estimate the visibility, prune the massive amounts of video so only images containing marine organisms remain, and finally classify the marine organisms. The hypotheses are that the visibility can be estimated based on 3D distance estimates of objects and abnormalities found by analyzing 3D motion patterns of image features as they move through the water. Lastly, the project investigates how automatic vs semi-automatic deep learning algorithms can be developed for this problem domain.