Congratulations to Shehan Amarasekera, studying BSc (Hons) Computer Science with a Year in Industry, who won the Most Innovative Solution prize, chosen by Jay Gildea of Fivium, at this year’s Computing Poster Fair.
Shehan’s project deals with analysing emotions to discover if facial analysis in a non real-time setting is as accurate as asking participants about their emotions, and if it could perhaps uncover any unconscious emotions. He aimed to determine which algorithm is more appropriate to use – Multi-Task Cascaded Convolutional Neural Networks (MTCNN) or Haar Cascades. Project participants played a segment of Super Mario Bros., during which their gameplay and face were recorded. Afterwards, they were be asked to fill out a questionnaire, and the facial capture was analysed by both algorithms. Then, Sheehan interviewed the participants to establish the accuracy of the analysis, taking into account the similarity of the algorithm’s results and participant’s questionnaire, and what the participants think after having seen the results.
Shehan found a few differences in the effectiveness of the two algorithms. He found that MTCNN is much more competent at face tracking, especially when obscured, whereas Haar occasionally assumes an object in the background is a face for a frame or two. However, Haar is much faster to run than MTCNN. Emotion recognition, on the other hand, features subtler differences between the two. While both are great at detecting happiness, MTCNN seems to detect sadness more often than Haar. Both algorithms can pick up anger, though their detection of it is much more subdued than happiness, for instance. Fear and disgust, on the other hand, are picked up on the least.
Shehan says: ‘I am beyond delighted to have won the Most Innovative Solution award for my project. It was a joy to talk about my project to all the people who visited. It has been so fun to work on it and I’m glad I could share my enthusiasm for it with everyone who came!’
Congratulations again to Shehan, and thank you to Jay Gildea of Fivium.