Cloudlet-based Large-scale 3D Reconstruction Using Real-time Data from Mobile Depth Cameras

Measuring the distance between observed objects and the camera, depth cameras on mobile devices are a leverage to more accurate and innovative vision-based applications. In this article, we present the initial design of a distributed cloudlet-based system that integrates depth maps crowd-sourced from mobile devices and head-mounted displays into a global 3D world model. To ensure fast enough processing of depth frames for real-time vision applications, the model is automatically split over multiple VMs when it becomes too large. By geographically distributing the VMs with submodels across cloudlets, our system provides the model as building block to low latency vision-based applications without overwhelming the network.