Overview / Usage

This is a Proof-Of-Concept (POC) repository for working through some of the intricancies of face detection and
emotion recognition. The basic strategy here is to work through some of the available options for pre-trained CNN's that
already do emotion recognition. The current implementation (#2 below) is accurate to ~60% or so. While this may seem low,
it is common for autistic children to be at 25% accuracy, and the consultants idea is to gameify the emotion detection.

Methodology / Approach

Leveraging mobile/mobile apps for assisting teachers of/caretakers of autistic persons, primarily focusing on children, we are working with experienced professionals to develop a mobile teaching tool. One aspect of this tool is automatic emotion detection via computer vision + machine learning.

Technologies Used

The main problems we are solving in this repo/POC are:

1. Obtain an emotion detection network
2. Use Apple Vision framework to detect faces
3. Run the detected face through a CNN ported to CoreML
4. Display the detected face and the top prediction

We use:

• Apple Vision framework
• Apple CoreML
• Currently harshsikka emotion detection pre-trained CNN, ported to CoreML but that is likely to change

While not part of the POC, this app will be developed for Android, in parallel.

Repository

https://github.com/revelfire/autism-emotion-aid/tree/master/Emotism/Emotism