Human-in-the-loop AAL Approach to Emotion Capture and Classification

In Ambient Assisted Living (AAL), emotion detection plays a crucial role in offering personalized and effective support to users. However, the lack of labeled data for emotion detection poses a significant challenge for developing intelligent solutions. To tackle this issue, we propose a human-in-the-loop (HITL) approach that engages users in acquiring and classifying their facial images, which can be utilized to augment existing emotion detection datasets. Our system is composed of three primary phases: facial image acquisition, emotion recognition model image classification, and user-based classification. By involving users in the data annotation process, we ensure the quality of the labeled data and also facilitate the collection of diverse and context-specific emotional expressions. We conducted experiments in a semi-supervised environment, where users catalogued frames according to the corresponding emotions, aiming to improve the model’s prediction accuracy for their emotions. The experiments supported the introduction of the user in the image classification process, promoting the creation of reliable labeled for later model training.