Approaches to 3 D Motion Estimation with Multiple Cameras ∗

The purpose of this article is to review the main approaches described in the literature to 3D motion estimation with multiple cameras. The review assumes that the initial 3D geometry of the scene has already been estimated using techniques such as laser-scanners [20], time-of-flight cameras [27], structured light measurements [44] or multiple view reconstruction [45]. This review concentrates on passive techniques that do not require manipulation of the scene, such as the addition of active LEDs markers or ‘ultraviolet paint’ [3], although some passive optical-marker techniques are included as a reference baseline. It is also important to note that this review focuses on recovering the dense motion of the scene, rather than tracking a single 3D object with methods such as [2, 22, 42]. The dense three-dimensional motion of the scene is often referred to as the scene flow [51], and is analogous to the concept of optical flow for describing the motions within a 2D image. There are several approaches to estimating scene flow: some ∗This article is taken from chapter 2 of my thesis [41] available at: www.dcs.warwick.ac.uk/∼tpopham/thesis.pdf