Efficient street sensor selection for traffic flow monitoring leveraging network observability metrics

For surveillance and safety purposes, the development of smart cities is currently focusing on traffic flow monitoring and congestion forecasting. These tasks frequently necessitate deploying a large network of sensors, but economic and environmental constraints may require starting with fewer devices. Network modeling in transport systems, which is crucial for selecting these sensors, often involves identifying supply, demand, and assignment models to capture network dynamics and parameters. Once a network model is available, sensor data can be gathered to refine traffic flow estimates. Additionally, this data allows for evaluating network conditions, such as vehicle travel times and traffic queue formations. This work addresses efficient sensor selection for monitoring urban roads. To determine the most effective sensor placement, model-based techniques leveraging linear time-invariant systems and network observability are commonly used. Central to these strategies is the observability Gramian, which measures the energy induced by the initial state's free response. Higher energy levels signify greater observability and reduced sensing costs. Some metrics from the observability Gramian are here tested to evaluate the quality of state estimation, e.g. including dimensions of the observable subspace, energy distribution, mode balance, eigenvalue representation, and ellipsoid volume. These indicators are crucial for assessing system performance, and a comparison among them is provided in this thesis. Lastly, the theoretical findings are validated through real-world numerical simulations, contributing to enhance traffic flow monitoring in the city of Padua, Italy.