Real-time foot traffic monitoring is now a key part of retail analytics, campus management, and smart surveillance. However, limitations in computing power make it hard to use heavy deep-learning models in low-power settings. This paper introduces a lightweight footfall counting system that uses YOLOv8n and YOLOv8s along with a centroid-based tracking method for effective ID persistence and directional counting. Experimental results indicate that YOLOv8n reaches 4.1 FPS on CPU-only systems with 98–99% ID stability, surpassing YOLOv8s in real-time performance. The system works well for embedded platforms, public monitoring, and budget-sensitive deployments.
Kotkar et al. (Thu,) studied this question.