CrowdLense: A Serverless, Privacy-Preserving Edge AI Framework for Sustainable Crowd-Sourced Real-Time Traffic Monitoring

Real-time traffic monitoring underpins intelligent urban mobility, enabling route optimization, congestion prediction, parking management, and emergency response. Existing centralized systems suffer from high bandwidth use, scalability limits, and privacy risks, while localized analytics lack global context. This work presents CrowdLense, a serverless, privacy-preserving, and crowd-sourced Edge AI framework for sustainable and low-cost traffic sensing. Deployed on Raspberry Pi devices with cameras, CrowdLense performs on-device object detection using a fine-tuned YOLO model trained on the Indian Driving Dataset (IDD) to extract privacy-safe metadata—vehicle type, count, timestamp, and node ID—transmitted via MQTT to a scalable cloud backend. The system integrates AWS IoT Core, Lambda, and DynamoDB with a React-based dashboard for real-time and historical visualization. Leveraging citizen participation, individuals deploy personal edge nodes that collectively contribute anonymized traffic intelligence, forming a decentralized and scalable sensing network without sharing raw video. Extensive benchmarking on Raspberry Pi 4/5 across multiple inference engines identifies YOLOv11-S + NCNN as the optimal configuration, balancing accuracy, latency, and efficiency. CrowdLense thus establishes a unified paradigm for sustainable, privacy-preserving, and crowd-sourced traffic intelligence.