edgeflow: AI-Driven Intersection Intelligence

This capstone project, a collaboration between 99P Labs and Electrical & Computer Engineering students at The Ohio State University, centers on designing an embedded system for real-time intersection mobility analysis. By integrating camera feeds, weather data from APIs, and advanced computer vision techniques (e.g., YOLO and DeepSORT), the team aims to streamline the detection of vehicles, track mobility patterns, and log relevant data for future insights. A key part of the design is a Raspberry Pi–based architecture bolstered by Coral Edge TPU hardware, ensuring rapid, on-device processing of video inputs for near-instant feedback.

‍

The system’s software-defined structure embraces modularity and edge computing principles. Mobility data is parsed locally on the Raspberry Pi to reduce latency, while a robust SQL database retains logs of vehicle counts, timestamped events, and weather conditions for historical analysis. Cloud services, built around AWS IoT integration, expand the system’s capabilities—allowing remote dashboard visualization, real-time alerts, and potential V2I (vehicle-to-infrastructure) communication scenarios. This hybrid approach balances performance demands with scalable data management, paving the way for sophisticated mobility applications such as dynamic signal control or predictive congestion modeling.

‍

Throughout this semester, the focus has been on designing the system’s architecture—finalizing hardware selections, implementing key software modules, and outlining data flow pathways from sensors to dashboards. The team validated early prototypes on local testbeds, exploring how best to integrate third-party libraries (e.g., Python-based YOLO frameworks, OpenWeather API) without overwhelming edge-device resources. With foundational elements now in place, the next semester will build upon these core modules, transitioning the project from design to implementation. During that phase, the team intends to deploy the system in a real or simulated intersection environment, gather performance metrics, and refine both hardware and software configurations to handle higher data throughput.

‍

Looking ahead, this ongoing initiative underscores the potential of low-cost, embedded solutions for advanced mobility management. By showcasing how edge computing and modular architectures can process complex real-time data streams, the project not only benefits academic exploration but also points to transformative, scalable solutions for cities grappling with modern mobility challenges. Next semester’s work will determine the system’s overall viability, setting the stage for future enhancements—such as integration with adaptive signal control algorithms or expansion to multi-intersection networks.