Spherical Panoramas for Pan-Tilt Camera Motion Compensation in Space-Variant Images

Abstract

In active vision scenarios, the motion of the observer induces an apparent motion in the image plane. One approach for camera motion compensation is the use of panoramic images, representing the scene at the different positions of the camera. In this work, an approach to build spherical panoramic views from a pan-tilt camera is described, which is based on background updating techniques. Interestingly, panoramic representations allow motion detection and analysis to be performed independently from camera movements. This makes easier the detection of moving objects in active tracking tasks. Additionally, the advantages of combining spherical panoramas with log-polar images is discussed. An example of target segmentation by background difference is shown, which shows the effectiveness of panoramic representations in active vision systems.