Inspiration
Missing scenes from netflix when I temporarily leave the laptop.
What it does
It uses deep learning to recognise the person via the laptop's webcam. After that, the daemon can be configured to do anything. For demo purposes, we are using it to pause netflix / youtube videos after 10 seconds. Another use case is to lock the laptop after a certain amount of time without the user sitting in front of the laptop. This is vital in a highly secure environment to avoid unauthorised access to data.
Other potential applications: We can see this being applied in areas where full attention of the user is required for high security reasons, such as in the following two use cases:
- Air traffic controllers work in a job that requires them to focus every second on the screen showing the current flow of air traffic. Even a few seconds can cause fatal plane accidents. Automatic Face Recognition can monitor in the background if the controller is focused on the screen, if not it will notify the supervisor.
- Baggage screening at airports requires the user to identify objects on the screen. As soon as Automatic Face Recognition realises that the user is not focused, it will automatically stop the belt and wait for the user to return.
How we built it
Used a popular open-source project, Boss Sensor as inspiration to use face recognition via deep learning, training on examples of faces. The deep learning is performed by using Tensor flow and Keras. To get the video feed, we're using OpenCV.
Challenges we ran into
-- Using Tensor flow and Keras for deep learning. A huge number of training examples are required. Therefore, it is not as accurate yet as we'd like it to be. -- Getting OpenCV installed on a mac. This was a nightmare.
Accomplishments that we're proud of
-- Able to differentiate between different faces as opposed to just traditional face detection. This is able to recognise different faces for example the face of an authorised user vs. an unauthorised user.
What we learned
We've learned how to use Tensor flow for deep learning. We also learned about feature recognition and image manipulation.
What's next for Automatic Face Recognition
We would like to automate the learning process of the algorithm. Currently, we have to manually capture images and train the algorithm. Automating this process would mean a larger training dataset, resulting in higher accuracy.
Log in or sign up for Devpost to join the conversation.