Jazmín.IA: Embedded Computer Vision System for the Analysis of Behavioral Indicators and Their Relationship with User Performance

Jazmín.IA is an embedded Edge AI system designed to investigate the relationship between observable facial behavioral indicators and user performance during digital tasks. The proposed approach focuses on measurable behavioral patterns rather than subjective emotion classification, using facial landmarks and geometric descriptors such as Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), blink frequency, eye-closure duration, and yawning-related events. The system is deployed on the Arduino UNO Q 4GB platform, enabling local video processing and reducing dependence on cloud services. MediaPipe Face Mesh is used for facial landmark extraction, while a preliminary convolutional neural network (CNN), initially developed using the FER-2013 dataset structure, serves as an auxiliary component during the early development stage. The CNN model was converted to TensorFlow Lite and reduced to approximately 443 KB for embedded deployment. Preliminary integration tests demonstrate local video acquisition, facial landmark extraction, EAR/MAR computation, temporal behavioral-event filtering, and on-device CNN inference. Importantly, these indicators are not interpreted as direct measurements of internal emotional or cognitive states; their association with fatigue, attention, cognitive workload, and user performance remains subject to experimental validation. Future work will evaluate processing performance and behavioral indicators across controlled user tasks and investigate Facial Action Units through the Action Decouple Multi-Tasking approach to obtain a richer representation of observable behavioral patterns.