Abstract:
Marine biodiversity serves not only as a major food resource but also give off various toxins that can be utilized in the development of medicines. Most of these species are endangered and if countermeasures are not taken they are going to be extinct soon. Currently, the research being done for the population estimation of these marine species is not only limited but it is also expensive and destructive. We want to develop a non-invasive and non-destructive method for fish biodiversity estimation using Computer Vision with supporting hardware to be deployed in fish habitat. To do so we have developed a software system which not only provides tools to label the collected data but also automatically detect and localize fish out of a video using Computer Vision techniques. In our way, we have developed a segmentation algorithm which gives better results than similar techniques applied to solve same problem on standard datasets.
