Occlusion Boundaries from Motion: Low-Level Detection and Mid-Level Reasoning

摘要: The boundaries of objects in an image are often considered a nuisance to be "handled" due the occlusion they exhibit. Since most, if not all, computer vision techniques aggregate information spatially within scene, spanning these boundaries, and therefore from different physical surfaces, is invariably erroneously together. In addition, convey important perceptual about 3D scene structure shape. Consequently, their identification can benefit many pursuits, low-level processing high-level reasoning tasks. While much focus placed on individual, static images, applications actually offer video, or sequences as input. extra temporal dimension data allows motion camera used processing. this paper, we exploitation subtle relative-motion cues present at boundaries. When combined with more standard appearance information, demonstrate cues' utility detecting locally. We also novel, mid-level model for globally object propagating such local extract improved, extended