Camera Image Synchronisation in Multiple Camera Real-Time 3D Reconstruction of Moving Humans

Abstract

We present an analysis of the requirement for input synchronisation in a multi-camera 3D reconstruction system for real-time applications such as telepresence.  Synchronisation of the cameras at the acquisition stage is universally used to ensure the images feeding the reconstruction algorithm were taken at the same time.  However, this requirement adds delays to the reconstruction pipeline, therefore increasing the end to end latency of the system.  While this has not been a significant problem for many of the applications of 3D reconstruction, it is for its application to tele-presence. Furthermore, synchronising the firing of cameras adds much financial cost to the system. Using real camera images of moving humans, we study the effect removing synchronisation has on the output reconstructed model over a range of camera configurations and relative frame delays.  From this we determine the synchronisation requirements for a 3D reconstruction telepresence system in terms of the maximum time between camera frames that gives rise to acceptable results.