Synthesizing test image sequences for omnicamera: translating and rotating spheres

Abstract

Omnidirectional cameras or omnicameras have been recently exploited in a variety of applications, including autonomous navigation, remote surveillance, video conferencing, and scene recovery. In these applications, there are important problems, such as motion detection, ego motion analysis, 3D motion inference, scene segmentation, and object structure or camera parameter analysis that require image flow from image sequences as a fundamental measure. Some synthetic image sequences are currently available for testing, but unfortunately, there is no algorithm for synthesizing image sequences with accurate groundtruth flow fields and motions close to reality for the omnidirectional camera. In this paper, we present an approach to synthesize test image sequences for omnidirectional cameras. Our algorithm provides synthetic test image sequences of translating and/or rotating spheres in front of a catadioptric omnicamera. With the known optic flow groundtruth, these synthetic image sequences can be used for evaluation and comparison among various image flow estimation techniques. The result image sequences of a translating sphere, a rotating sphere, and both translating and rotating sphere are generated. Their central frame and groundtruth flow fields are also shown.