Allec does not just record the world around him, he actually sees it.
To be able to do this, Allec’s brain contain his own version of computer vision.
Allec is able to categorize what he sees into 10 categories, rating the likeliness of what he saw with at 0 to 100 % likeliness, for each category.
The categories as of now are:
- Sea weed
- Wreck
- Diver
- Fish
- Jellyfish
- Open water
- Swimmer
- Allec
- Fishing net
- Sea bed
About every two second Allec categorizes a picture taken, rating the picture for each of the 10 categories.
He will pass that information on to the AI system, that can then react in an appropriate manner.
Sometimes it’s nice to know something a bit more detailed. For this Allec is able to learn 3 additional sets of categories. The idea behind this is that if say Allec spots a fish, he can switch to the more fish detailed secondary category set and learn that the fish seen was a shark. The additional sets of categories are also 10 categories each.
The PC user interface allows you to select other categories to look for, replacing one or more of the above, in both the primary and the secondary category sets. Some of them (Sea weed, Open water, Fishing net and Sea bed) are really required for Allec to be able to find his way around, so cannot be replaced.
To give an example of the power of computer vision, we have this little history from very early in the development of the computer vision part. At that time we were using a set of pictures for testing, with categories like dog and horse, but no cows. Of course we had to show the system a picture of a cow to see what would happen. Well, the system identified the black and white cow as equally likely a dog and a horse. The horse because the general shape with four legs and the dog because the black and white pattern of the cow, was also seen on some dogs. Based on the available facts, the computer vision system came with the best solution possible.