Multimodal Group Emotion Recognition In-the-Wild Towards a Privacy-Safe Non-Individual Approach
This thesis introduces novel frameworks for group emotion recognition in real-world scenarios, prioritizing privacy by analyzing collective audio-video signals rather than individual cues. The proposed cross-attention multimodal architecture with Frames Attention Pooling (FAP) and a Variational Encoder Multi-Decoder (VE-MD) framework demonstrate competitive performance without relying on individual facial or vocal data. These contributions aim to advance affective computing by enabling privacy-safe group emotion analysis. AI
IMPACT Introduces new methods for privacy-preserving affective computing, potentially enabling broader adoption of emotion recognition in sensitive group contexts.