This research demonstrates how large vision language models can recognize emotions beyond facial expressions by incorporating contextual cues, body language, and commonsense reasoning.

• Explores two major approaches: image captioning with language-only models and direct vision-language analysis
• Addresses limitations of traditional facial expression analysis by incorporating full-body cues and environmental context
• Achieves more nuanced emotion detection by mimicking human emotional theory of mind capabilities
• Enhances security applications through improved detection of distress signals and unusual emotional states

This breakthrough enables more robust security monitoring systems that can identify emotional distress in varied contexts, potentially improving threat assessment and emergency response.

Contextual Emotion Recognition using Large Vision Language Models