Architectures for Computer Vision: From Algorithm to Chip with Verilog

7 Motion and Vision Modules

This chapter introduces some of the issues associated with vision modules and their integration. For the issue of motion, we first review the geometry of the 3D and 2D motion fields. We also review the basics of structure from motion, and then focus on optical flow, in which we examine various constraints in a more fundamental manner and review basic energy minimization. The contents are not intended to be complete in scope and depth, as the aim is to lay the groundwork for the topics in subsequent chapters. For a more comprehensive treatment of motion, please refer to books on multiple view geometry (Faugeras and Luong 2004; Hartley and Zisserman 2004), reviews of optical flow (Baker et al. 2011; Barron et al. 1994; Fleet and Weiss 2006; McCane et al. 2001), and scene flow (Cech et al. 2011; Huguet and Devernay 2007; Vedula et al. 2005). In addition, a web site dealing with optical flow is also available (Middlebury 2013).

Like binocular stereo vision, motion vision is one of the major vision modules by which we can induce motion information in addition to the depth of the surface shape and the volume information of objects. Created by a camera, the successive frames of images contain depth information by means of optical flow. The relative velocity between the scene motion and the egomotion appears as the motion field, when it is projected onto the image plane. The optical flow refers to the apparent velocity of the motion velocity when viewed ...