A Survey on Privacy-Preserving Emotion and Attention Recognition for Children with ASD
Year: 2026
Type: conference-paper
Study design: Systematic Review
Abstract: The application of Computer Vision (CV) to support scalable clinical interventions for Autism Spectrum Disorder (ASD) has grown significantly, presenting a critical opportunity for global public health. However, the field faces major challenges regarding the privacy of sensitive data and the poor generalization of models trained on neurotypical individuals. This systematic literature review maps the state-of-the-art in emotional and attentional analysis technologies for ASD, with a specific focus on privacy-preserving, geometry-based approaches (facial landmarks and gaze tracking). Analyzing studies from 2011 to 2025, we identify a clear methodological shift from classical geometric classifiers to Deep Learning architectures, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Graph Neural Networks (GNNs). The findings indicate that fusing high-precision gaze tracking with facial geometry significantly improves detection accuracy. Additionally, to overcome the scarcity of public, ASD-specific datasets caused by ethical restrictions on raw video sharing, we conclude that transitioning to abstract geometric representations is essential. Ultimately, this methodology offers a pathway to lightweight, real-time AI models optimized for edge devices, establishing a generalized, privacy-preserving framework.
Institution: Instituto de Pesquisas Eldorado
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: hybrid
OpenAlex ID: W7203840235
