This research introduces EALD-MLLM, a novel multi-modal framework that analyzes emotions in long-duration videos while protecting privacy through de-identification techniques.

• Addresses critical limitations in current emotion AI by focusing on long sequential videos that capture authentic emotional states rather than just momentary reactions
• Employs de-identification techniques that preserve emotional signals while removing personally identifiable information
• Leverages multi-modal large language models to process both visual and textual information for comprehensive emotion analysis
• Demonstrates superior performance in emotion recognition while maintaining privacy standards

This advancement has significant implications for security and privacy applications, enabling emotion-aware systems that can function without compromising user identities or storing sensitive biometric data.

EALD-MLLM: Emotion Analysis in Long-sequential and De-identity videos with Multi-modal Large Language Model