A Lightweight Method for Pedestrian Flow Analysis in Public Spaces

This paper presents a lightweight pedestrian flow analysis system based on the NanoDet model to reduce the labor costs of traditional monitoring. By processing video streams in real time, the system performs pedestrian detection, tracking, and counting using inter-frame displacement and stores results (count, position, time) in a MySQL database. A visualization interface displays detection outcomes. Experiments show that, compared to YOLO-based methods, the proposed approach offers faster processing and higher mAP, meeting pedestrian flow analysis needs in crowded scenes.