Super-resolution omnidirectional camera images using spatio-temporal analysis

Recent years have witnessed the increasing popularity of studies using 360° images captured with omnidirectional cameras. The authors have already created a database of building images using an omnidirectional camera. Omnidirectional cameras can capture 360° images in a single shot, but their optical characteristics are different from those of ordinary cameras, and their use presents several problems. Two problems in particular pertain to camera calibration and the low resolution obtained when capturing a single 360° image. Numerous studies have been conducted in regard to the former, but only a few in regard to the latter due to the small number of instances in which omnidirectional camera images have been used. This has come to be regarded as an especially significant problem, as the number of opportunities for omnidirectional cameras to actually be used has been increasing of late. Resolution-enhancement techniques can be thought of as being broadly divided into hardware methods, in which a high-resolution CCD or the like is employed; and software methods, in which super-resolution or the like is performed using a series of multiple images. In this report, we discuss the use of software-driven methods to super-resolve and otherwise convert low-resolution omnidirectional camera images. Images that have been examined with conventional super-resolution techniques have hitherto been almost exclusively taken with static cameras, whereas we have adopted a novel approach in using images captured with a moving camera. Moreover, our proposed technique is applicable not only to omnidirectional cameras, but to ordinary image capture as well.