A Multimodal Fusion Framework for Enhanced Exercise Quantification Integrating RFID and Computer Vision

The emerging paradigm of embodied intelligence, which emphasizes the tight coupling between physical embodiment and cognitive processes, has opened up new frontiers in Human-Computer Interaction (HCI), particularly for smart personalized exercise monitoring. Despite its significant potential, existing systems often struggle in multi-user environments, where the accuracy and reliability of exercise tracking are severely compromised by the complexity of human behaviors and interactions. To address this critical issue, this paper proposes a multimodal fusion framework that integrates Radio Frequency Identification (RFID) and Computer Vision (CV) technologies for personalized exercise monitoring in multi-user scenarios, which is the first work to enhance interaction-aware perception in such settings. The system workflow consists of three core modules: Data acquisition, perception modeling, and exercise monitoring, which collectively enable comprehensive analysis of individual exercise behaviors. The system leverages multimodal data from RFID tags and a depth camera to jointly identify object ownership and recognize user activities in real time. Extensive experimental results demonstrate that the proposed system achieves an average matching accuracy of 95%, and an average estimation accuracy of 94%.