We are progressing! We have trained the machine to detect dolphins as dolphins and we applied a tracking algorithm to this object detection output. Check it out!
We trained an object detection model on a a machine learning platform, box21, to detect birds, boats, and people swimming.
We noticed that other objects in the same image, most often birds, were automatically classified as "dolphins" in our model. This means that our model did not know the difference between different objects. To track dolphins, we had to further train the machine categorizing new photos with dolphins and birds. The final object detection looks like this:
Since then, we have been able to apply the tracking algorithm to dolphins.
Cool huh?
We are presenting the preliminary results in a poster at the European Cetacean Conference in Catania, Italy!
Author: Fadia Al Abbar
