Скачать или смотреть Use of AI in Fish Counting

Technology is only as good as it is useful. Sometimes we can be more focused on hype than on helpfulness.

One of the first projects I have the privilege of working on in Metropolitan Wireless Internationla (MWI) is to help fish farmers leverage technology to track their count of fish.

How many fishes do they have at any moment in time? This information is important for at least 2 reasons. Firstly, it will determine the health trajectory of the fishes. Fishes that don't grow well for whatever reasons will die. You don't want to be surprised only at harvest time. And more importantly, it improves accuracy of how many fishes a farm can expect to supply to their customers (supermarkets, restaurants, consumers) at harvest time.

But there is a catch. Like human beings, stress affects the growth development of fish, and may even cause some of them to die prematurely. The tracking solution must be "non-intrusive" and ride along with the unforced rhymes of the fish's natural movements.

Together with the team in MWI, we came out with a computer vision AI-based fish counting solution. The solution makes use of only videos taken top down of the fish (completely 'non-intrusive').

There are 3 scenarios where we can deploy this AI solution.

1. Object detection - when the fishes are corralled in a net

2. Object tracking - when the fishes swim from point A to point B at feeding time

3. Object detection at harvest time

The object here may be fishes. But the use cases extend beyond fish. Consider how object detection and tracking can be used to identify and track movement of people for purpose of crowd control, or to study human traffic pattern on the subway or a mall.

The solution is hosted on the cloud on AWS sagemaker studio so as we gather more videos, we will continue to improve prediction accuracy further. Enjoy the videos.