Understanding how people and vehicles move through public spaces is essential for designing inclusive, safe, and efficient urban spaces. In the Mendrisio district (Switzerland), we deployed a low-cost, AI-powered sensor to monitor pedestrian and vehicle flows in different seasons. The sensor uses camera-based image recognition to detect and classify objects in real time while preserving privacy by avoiding biometric or identity-related data capture, in full compliance with GDPR. The system was developed using open source hardware and software. Processes video frames on edge using a lightweight machine learning model optimized for embedded devices and periodically transmits summary data (object type, direction, timestamp) via NB-IoT to a centralized data platform. The collected data was used to generate temporal analyses and heatmaps of space usage and to validate the classification accuracy in different weather and lighting conditions. Field tests demonstrated the sensor’s capability to operate autonomously for extended periods with low power consumption, while highlighting limitations in NB-IoT connectivity in specific locations. Despite these constraints, the system provided valuable information on public space utilization, identifying peak hours and spatial patterns relevant for mobility planning and urban design. This approach offers a replicable and cost-effective solution for municipalities seeking data-driven support for decision-making. By combining privacy-sensitive AI, open technologies, and standard data models (such as SensorThings API), the project contributes to a more transparent and inclusive digital urban ecosystem.