Edge-Optimized YOLO Model for Active Mobility Detection in Citizen-Led Urban Analytics

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Urban mobility monitoring requires accurate, real-time object detection under the computational constraints of edge devices. This study develops a lightweight deep learning model to support Citizen-led Automated Modal Infrastructure Analytics (CAMINA), a community-driven approach that emphasizes active transportation modes. We present an edge-optimized object detection pipeline based on YOLO11n, the nano variant of YOLO11, designed for low-power deployment. The model detects nine road user classes: person, cyclist, e-scooter, car, SUV, motorcycle, bus, delivery van, and truck. A hybrid labeling strategy combining YOLO11 (for COCO classes) and YOLO-World (for open-vocabulary detection) was applied to annotate 1,895 urban images with semi-automated assistance and manual correction. Comparative experiments across YOLO variants (YOLOv5n, YOLOv8n, YOLO11n) show that our optimized YOLO11n achieves 0.56 mAP@0.5 at 15 fps on a Raspberry Pi 5 (8GB RAM) with a size of 10 MB. These results demonstrate practical feasibility for distributed, citizen-led monitoring using low-cost devices. By combining edge-optimized performance with a community-centered deployment model, CAMINA offers a scalable and privacy-preserving approach to urban mobility analytics. While broader validation across diverse city contexts remains necessary, this work highlights the potential of citizen-led sensing to complement traditional traffic monitoring and support active mobility planning.