Real-time shading-based refinement for consumer depth cameras

摘要: We present the first real-time method for refinement of depth data using shape-from-shading in general uncontrolled scenes. Per frame, our algorithm takes raw noisy and an aligned RGB image as input, approximates time-varying incident lighting, which is then used geometry refinement. This leads to dramatically enhanced maps at 30Hz. Our makes few scene assumptions, handling arbitrary objects even under motion. To enable this type map enhancement, we contribute a new highly parallel that reformulates inverse rendering optimization problem prior work, allowing us estimate lighting shape temporally coherent way video frame-rates. minimized regular grid Gauss-Newton solver implemented fully on GPU. demonstrate results showing maps, are comparable offline methods but computed orders magnitude faster, well baseline comparisons with online filtering-based methods. conclude applications higher quality improved surface reconstruction performance capture.