An End-to-End Edge-Computing Framework for IoT-enabled Monitoring

The exponential growth of Internet of Things (IoT) applications across diverse domains, from healthcare and agriculture to automated driving, demands innovative solutions for efficient, scalable, and application-independent monitoring. This thesis introduces an end-to-end edge-computing framework designed to address the limitations of current IoT monitoring systems, which often require costly, customized implementations for specific applications. By combining modular edge processing capabilities with deep learning (DL) optimizations, this framework enables real-time data processing directly on edge devices, enhancing adaptability and performance across varied use cases. The proposed framework addresses three critical areas of IoT monitoring: project management and reporting, optimized DL-based prediction on the edge, and efficient dataset preparation. First, the project monitoring module leverages the open-source Measurify platform to provide users an effective way to store, manage, and retrieve progress indicators (PIs). These capabilities are validated in the Hi-Drive project, a European initiative focusing on expanding the operational design domain for automated driving systems. Second, the framework includes a neural architecture search (NAS) component tailored for resource-constrained edge devices, with specific enhancements for time-series prediction tasks. This component is tested on a structural health monitoring case study, demonstrating effective DL deployment on resource-constrained hardware without sacrificing accuracy. Finally, the dataset preparation module introduces a streamlined, open-source workflow for collecting and processing labelled data across multiple domains, validated through a sports activity monitoring use case. Results from these applications confirm that the framework can support IoT-enabled monitoring with lower latency, reduced cloud dependence, and greater flexibility than existing solutions. This work lays the foundation for further development of adaptable, scalable, and cost-effective IoT frameworks capable of meeting the demands of diverse, data-driven environments.