Depth sensing using active coherent illumination

We examine the use of active coherent sensing-an increasingly available technology-for sensing the depth of scenes. A scene is a sparse signal but also exhibits significant structure which cannot be exploited using standard sparse recovery algorithms. Instead, inspired by the model-based compressive sensing literature we develop a scene model that incorporates occlusion constraints in recovering the depth map. Our model is computationally tractable; we develop a variation of the well-known model-based Compressive Sampling Matching Pursuit (CoSaMP) algorithm, and we demonstrate that our approach significantly improves reconstruction performance.