PhD in AI and Computer Vision for Emotion Understanding in Videos Focused on ASD Interaction Analysis

Project Overview

This PhD project aims to innovate in facial expression recognition (FER) and emotion understanding in videos, focusing on the analysis of adult-child interactions, particularly related to the neurodevelopmental monitoring of children with Autism Spectrum Disorder (ASD). Existing FER systems are often trained on exaggerated emotions from social media or movie datasets, which do not represent everyday human expressions, especially subtle emotions seen in clinical settings. This project will develop deep neural network models capable of recognizing both typical and nuanced emotional expression patterns in unconstrained video scenes, accounting for challenges such as varying illumination, head pose, and camera distance.

What You Will Do

The candidate will explore architectural and training strategies to enhance the robustness of FER methods, adapting multimodal and visual foundation models to domain-generalize across generic datasets and specialized healthcare contexts. You will investigate recognition of micro-expressions relevant for ASD diagnosis, such as subtle eye blinks, frowns, and smiles. This involves developing algorithms that perform reliably under occlusions, low facial visibility, and varying camera placements common to medical consultations.

Expected Outcomes

The expected outcome is a set of deep learning models that improve automated recognition and interpretation of emotions in children with ASD during interactions with caregivers, aiding clinicians with objective, timely data. You will contribute to the international ANR project AIMAINT and disseminate research findings at major AI and computer vision conferences, facilitating translation to practical diagnostic and social analysis tools.

Why This Matters

Enhancing FER for ASD evaluation addresses significant gaps in current diagnosis practices, which are subjective and time-consuming, by providing objective, automated behavioral coding tools. The research advances AI's capability to understand complex social signals in real-world scenarios, impacting security, healthcare, and social sciences by enabling better monitoring and understanding of human emotional behaviors.