Educational Supplement to Research Initiaion Award: MachineVision

点击翻译按钮获取中文摘要

Points in scene, when illuminated, reflect light not only in the direction of the sensor but also between themselves. These interreflections can appreciably alter the appearance of the scene. Existing vision sensors and algorithms do not account for the effects of interreflections and hence often produce erroneous results. An initial investigation studied the effects of interreflections on shape information recovered by shape-from- intensity methods (e.g. photometric stereo). It was discovered that, for Lambertian surfaces, the erroneous (pseudo) shape recovered in the presence of interreflections is unique and can be mathematically related to the actual shape. An algorithm has been developed that uses a physical interreflection model to recover the actual shape of the surface from the psuedo shape. The research will be extended in the following directions. (1) The analysis of non-Lambertian (e.g. specular) interreflections. This entails the development of mathematical models for non-Lambertian interreflections as well as their application to the problem of recovering accurate shape in the presence of interreflections. (2) The detection and correction of errors produced by existing vision sensors (spot scanners, light stripe range finders, stereo systems) as a result of interreflections. For each sensor, a sensor model is used to obtain physical constraints that aid the analysis of interreflections. All analytical results will be verified by laboratory experiments. This research should achieve (a) fundamental advances in the analysis of interreflections and (b) substantial improvements in vision sensing technology.