Accidental cameras : creating images from shadows

Abstract

In this thesis, we explore new imaging systems that arise from everyday occlusions and shadows. By modeling the structured shadows created by various occlusions, we are able to recover hidden scenes. We explore three such imaging systems. In the first, we use a wall corner to recover one-dimensional motion in the hidden scene behind the corner. We show experimental results using this method in several natural environments. We also extend this method to be used in other applications, such as for automotive collision avoidance systems. In the second, we use doorways and spheres to recover two-dimensional images of a hidden scene behind the occlusions. We show experimental results of this method in simulations and in natural environments. Finally, we present how to extend this approach to infer a 4D light field of a hidden scene from 2D shadows cast by a known occluder. Using the shadow cast by a real house plant, we are able to recover low resolution light fields with different levels of texture and parallax complexity.