Distributed volumetric scene geometry reconstruction with a network of distributed smart cameras

Abstract

Central to many problems in scene understanding based on using a network of tens, hundreds 1. An Overview of the System MotivationWith the recent development of cheap and powerful visual sensors, wireless chips and embedded systems, cameras have enough computing power to do some on-board "smart" processing. These "smart cameras" can form a network to collaboratively monitor, track and analyze the scenes of interest. This area have drawn a lot of attention in both academia and industry over the past years (see [1] [11] and [13] for an overview). However compared with the maturity and availability of the camera network hardware, the software capable of fully utilizing the huge amount of visual information is greatly under-developed. This has become the bottleneck for the wide deployment of the smart camera network (also called visual sensor network, VSN). There is an obvious demand to synchronize the recent development of vision algorithms with the development of the visual sensor network hardware. Our paper presents a completely new and natural approach to 3D reconstruction within a smart camera network.

Affinity-based distributed algorithm for 3D reconstruction in large scale Visual Sensor Networks

“…V, this approach exploits the spatial clustering of typical real data [32], [1], [14], allowing a significant reduction of the reconstruction computational time. Different strategies to distribute the computation burden over a camera network have been considered in [3], [5], [6], [15].…”

A structure-based approach for optimizing distributed reconstruction in Motion Capture systems

“…This task is practically unfeasible when m and n are large and by using a centralized management of the information, where computations are performed by means of a single computational unit. Instead, as commonly done in recent works based on smart cameras , Jovanovic et al [2006], Liu et al [2009], Klausner et al [2008], Aghajan et al [2008]), here each camera is assumed to be provided with computational and memory resources: the first goal of the paper is that of proposing an efficient strategy to distribute the computation among the available computational resources.…”