Abstract:

Cheap cameras and fast processors have made it possible to visually exploit geometric constraints in real time. It has been shown that the fast depth segmentation (FDS) algorithm successfully exploits geometric constraints to perform visual foreground/background segmentation in environments where other vision routines fail. This paper presents new insights into the operation of the FDS algorithm that lead to the concept of a virtual surface margin. We then show how surface margins can be used to extend the FDS algorithm and thereby enable a class of logical volume operations that go far beyond simple background segmentation tasks. An example application called TouchIt is demonstrated. Touchit utilizes surface margins to create a virtual volume configuration that is useful for detecting physical proximity to a surface. We also present refinements in the the implementation of the FDS algorithm that make these volumetric computations practical for interactive applications by taking advantage of the single instruction, multiple data (SIMD) instruction set extensions that have recently become commonly available in consumer-grade microprocessors.