| Synthetic aperture imaging is a technique that projects images of a scene from different views on to a virtual focal surface, enabling us to "see through" occlusions in the scene. Using a 100-camera array we have used synthetic aperture imaging to view objects concealed behind dense foliage and to track a person moving through a crowd. The ability to see through occlusions makes synthetic aperture imaging a potentially powerful tool for surveillance.; This work makes two contributions. First, we characterize the image warps required for synthetic aperture imaging using projective geometry. This analysis leads to a robust camera calibration procedure for synthetic aperture imaging. Our analysis also shows the relation between the geometric complexity of the image warps and camera/focal plane configurations. In particular, we show that we can vary the focus through families of frontoparallel and tilted focal planes by simply shifting and adding the camera images. As image shifts are relatively simple to realize in hardware, this leads to a real-time system, which we demonstrate for tracking a person moving through a crowd.; Second, we explore methods to achieve sharp focus at every pixel in the synthetic aperture image by reconstructing the 3D surfaces of the occluded objects we wish to see. We compare classical shape from stereo with shape from synthetic aperture focus, and describe variants of stereo that improves image contrast by deleting some of the light rays that are incident on the occluder. |
