Pedestrian Eyes: An AI-Powered Framework for Real-Time Pedestrian Detection and Safety Analytics in Dhaka

Abstract

The pedestrian environment in rapidly urbanizing cities such as Dhaka presents acute safety challenges due to high densities, informal crossings, encroached sidewalks and limited monitoring infrastructure. This paper presents a comprehensive system that integrates computer vision (pedestrian detection, multi-object tracking, re-identification and event detection) with an NLP-powered public-sentiment analysis module, targeted at pedestrian safety and flow monitoring in Dhaka. A lightweight detector is fine-tuned for local conditions, a tracker and re-id pipeline supports cross-camera flow analytics, and a public-sentiment module mines social media and local news to prioritize intervention zones. The system is designed for edge-server hybrid deployment, emphasizes privacypreserving data handling and produces policy-relevant dashboards (counts, heat-maps, alerts). Preliminary experiments on urban footage show detection precision of ~85 %, tracking IDF1 = 72 %, and sentiment analysis accuracy of 78 %, demonstrating viability for municipal deployment and infrastructure planning within Dhaka city.