Research and train a Machine Learning Algorithm for Pain Recognition through Facial Expressions

Description:

My role in the Team was to focus on finding, training and testing the algorithm

I was part of a team. My role was to find and train the algorithm and ensure it worked properly. My colleagues focused on important ethical, historical, and presentational aspects related to pain perception.

Datasets

I used public databases of facial expressions categorized by emotions and Python to train the algorithm

The initial results

The results were promising, so I extended the research to pain expression recognition, which is said to be the most complex facial expression to recognize because it can resemble anger or sadness. Eventually, the algorithm began to become confused by the pain expression training.

A problem of Dirty Data?

At the beginning I thought the problem was the data I was training the algorithm with. So, I started to look deeper into the database and the images it contained. I found that the people in the dataset were in pain, but when I compared them to the people in the anger category, I was the first to have difficulty distinguishing between the two.

So, I began searching online for a solution and discovered that this is a common problem in emotional recognition algorithms. So, I took the opportunity to explain this common tough challenge. The stakeholders were satisfied with this exciting development in the research.

Conclusions

My research was successful. The algorithm could recognize the basic emotions in human expressions. The extension training I tried afterward revealed that pain is a complex expression that can be mistaken for many others. I am proud of this work because I went beyond the scope of the research and discovered some exciting challenges that facial expression recognition algorithms still face.

Downloads:

Final presentation

My contribution to the research project

Try yourself the code (without pain recognition)

Try yourself the code (with pain recognition)

Version:

1.0

Date:

22 Dec 2021

Client:

University of Milan

Project Duration:

1 month – part-time

Place:

Milan, Italy

Language:

Italian

Team:

3 students