A Decentralized Digital Twin via Crowdsourced Sensing and Browser-Based Edge Computation

Sean Hardesty Lewis, Junfeng Jiao, Yiming Xu, Jihyung Park, Connor Phillips

Keywords: digital twin, edge computing, crowdsourced sensing, browser-based computation, privacy-preserving AI, smart cities

TL;DR: A browser-based digital-twin system that fuses crowdsourced webcam detections in real time, approaching centralized accuracy while cutting bandwidth and cost by over 20×.

Abstract: Digital twins promise to revolutionize the management of complex urban systems by enabling real-time monitoring, prediction, and control. Existing platforms, however, often rely on dense deployments of calibrated sensors and centralized compute infrastructure, which limits scalability and accessibility. We introduce StreamTwin, a decentralized digital-twin framework that treats publicly accessible webcams as sensors and uses the web browsers of viewers as opportunistic edge-computing nodes. Object detections produced on client devices are fused into a coherent world model by our Aggregate Spatiotemporal Cache (ASC) algorithm. This enables interactive visualization of traffic conditions without ever transmitting raw video off the client, reducing deployment cost and network load while inherently preserving privacy. We detail the system design, data-fusion pipeline, implementation, and evaluation. Experiments on ten live traffic cameras show that StreamTwin reconstructs scenes with 0.73 IoU, approaching centralized baselines, while reducing per-stream bandwidth from 5 Mbps to 20 kbps. This reduces monthly operating costs by more than $20\times$. By removing specialized hardware requirements and supporting crowd participation at a global scale, StreamTwin lowers the cost and technical barriers to deploying digital twins.

Submission Number: 51