On the Performance Analysis of Active Visual 3D Reconstruction in Multi-agent Networks

Abstract

Visual 3D reconstruction builds the 3D map of the environment from images and is essential in a wide range of applications such as robotics, augmented reality and relic preservation. In this paper, we integrate the visual 3D reconstruction with the mobility of mobile platforms to address the active visual 3D reconstruction problem in multi-agent networks. We first establish a statistical model of the visual 3D reconstruction problem in multi-agent networks. Then we propose a next-best-view selection scheme to find the best camera configuration in active reconstruction. Moreover, we propose a statistical evaluation criterion to substitute traditional laser-scanned-model-based methods to measure the reconstruction quality under certain camera configuration. Numerical results verify the effectiveness of our methods.