New AI Framework Measures Student Engagement via Video Analysis
Researchers have developed a three-stage framework for video-based student engagement measurement that leverages vision-language models and large language models. The approach addresses privacy concerns by requiring only a few training samples for action recognition, using a sliding temporal window to segment 2-minute videos into non-overlapping segments. The framework incorporates peer context, which previous methods ignored, by analyzing students' actions in relation to classmates. This work was published on arXiv under ID 2601.06394.
Key facts
- Framework uses vision-language model for student action recognition with few-shot adaptation
- Sliding temporal window technique divides 2-minute videos into non-overlapping segments
- Incorporates peer context by analyzing classmates' actions
- Published on arXiv with ID 2601.06394
- Addresses privacy concerns by reducing need for large annotated datasets
Entities
Institutions
- arXiv