Research on a Multimodal Classroom Engagement Data Analysis System

In response to the strong subjectivity and quantification challenges inherent in traditional teaching evaluation systems, this research designs an auxiliary system for assessing teaching effectiveness based on multimodal perception. The system utilizes computer vision and artificial intelligence technologies to capture and analyze students’ multi-dimensional behavioral features—including head pose, eye state, mouth action, and facial expressions—in real-time via cameras. This data constructs a dynamic classroom engagement evaluation model. Relying on an STM32 embedded architecture and the MediaPipe framework for edge computing, the system ultimately outputs visualized engagement curves and, innovatively, generates teaching optimization suggestions using a Large Language Model (LLM). This provides teachers with actionable data support to adjust their teaching strategies, fully embodying the informatized characteristics of the smart classroom.