Dynamic Rendering of Remote Indoor Environments Using Real-Time Point Cloud Data

Abstract

Modern color and depth (RGB-D) sensing systems are capable of reconstructing convincing virtual representations of real world environments. These virtual reconstructions can be used as the foundation for virtual reality (VR) and augmented reality environments due to their high-quality visualizations. However, a main limitation of modern virtual reconstruction methods is the time it takes to incorporate new data and update the virtual reconstruction. This delay prevents the reconstruction from accurately rendering dynamic objects or portions of the environment (like an engineer performing an inspection of a machinery or laboratory space). The authors propose a multisensor method to dynamically capture objects in an indoor environment. The method automatically aligns the sensors using modern image homography techniques, leverages graphics processing units (GPUs) to process the large number of independent RGB-D data points, and renders them in real time. Incorporating and aligning multiple sensors allows a larger area to be captured from multiple angles, providing a more complete virtual representation of the physical space. Performing processing on GPU's leverages the large number of processing cores available to minimize the delay between data capture and rendering. A case study using commodity RGB-D sensors, computing hardware, and standard transmission control protocol internet connections is presented to demonstrate the viability of the proposed method.