Facial Focus Recognition

Inspiration

I was inspired by the prize tracks to think about issues I had and/or friends had in relation to education. This prompted me to think about some issues I heard people complaining about and thought about what some root common causes may be. I came to the conclusion that the problems arose from lack of attention/boring lessons.

What it does

My program is meant to try and assist teachers in figuring out what lessons work best for their class, as well as maybe figure out which students may need help focusing based on the analysis the project would give the teacher. For example if someone was constantly looking down, then they are either tired or always on the phone, so a phone "locker might be useful for that student.

How we built it

I built this program solo. I had an interest in AI and the like for a long time. I thought about how to go about this and decided to use python as the language since I know it had useful libraries for AI related projects and went from there. Tutorials, mixed with reading documentation I ended up using 2 cascade files in order to detect faces from live captured faces and then use python libraries to predict the direction based on the model that was produced from the dataset gotten online and putting it through the code I made with help from tutorials.

Challenges we ran into

I ran into challenges mostly due to time limitations and not being able to find a good and already sorted dataset, which made me create code to sort the downloaded dataset as best as possible. The dataset also appeared incomplete from what they said online. It also had weird angles and limited angles, but a lot of pictures which is why I decided to use it since I knew that the training to make a useful model would take a long time. I also decided to use the entire dataset to train a model to see how consistent the dataset is, and make sure it is learnable. I used the dataset only for training to tune the model-hyper-parameters.

Accomplishments that we're proud of

I am proud that it functions well in real time as well as the fact that for the clear directions it is usually right. But because of the weird angles some of its guesses are off.

What we learned

I learned that the main focus and issues with this type of production is being able to detect faces well as well as Actually acquiring a good and useful dataset. the rest though might take some time and effort the first time. It becomes easier the next time.

What's next for Facial Focus Recognition

The next step would be to try and find a better way to detect faces. The thing after that would be to possibly find a better dataset(Images) or resort the dataset manually in order to get more accurate predictions. I didn't want to do that for time as well as while researching it I briefly saw something about how that would be missing it up somehow, so until I look more into it, I decided to just use the dataset as is after sorting it based on its naming convention. I also plan on trying to retrain the model after splitting the dataset into training, validation, and testing.

Built With

• cv2
• multi-pie
• python
• tensorflow
• vscode