Editing Motion capture (section)

====RGB-D cameras====
RGB-D cameras such as [[Kinect]] capture both the color and depth images. By fusing the two images, 3D colored [[voxels]] can be captured, allowing motion capture of 3D human motion and human surface in real-time.

Because of the use of a single-view camera, motions captured are usually noisy. Machine learning techniques have been proposed to automatically reconstruct such noisy motions into higher quality ones, using methods such as [[lazy learning]]<ref>{{cite journal |last1=Shum |first1=Hubert P. H. |last2=Ho |first2=Edmond S. L. |last3=Jiang |first3=Yang |last4=Takagi |first4=Shu |title=Real-Time Posture Reconstruction for Microsoft Kinect |journal=IEEE Transactions on Cybernetics |date=2013 |volume=43 |issue=5 |pages=1357–1369 |doi=10.1109/TCYB.2013.2275945|pmid=23981562 |s2cid=14124193 }}</ref> and [[Gaussian]] models.<ref>{{cite journal |last1=Liu |first1=Zhiguang |last2=Zhou |first2=Liuyang |last3=Leung |first3=Howard |last4=Shum |first4=Hubert P. H. |title=Kinect Posture Reconstruction based on a Local Mixture of Gaussian Process Models |journal=IEEE Transactions on Visualization and Computer Graphics |date=2016 |volume=22 |issue=11 |pages=2437–2450 |doi=10.1109/TVCG.2015.2510000|pmid=26701789 |s2cid=216076607 |url=http://nrl.northumbria.ac.uk/id/eprint/25559/1/07360215.pdf }}</ref> Such method generates accurate enough motion for serious applications like ergonomic assessment.<ref>{{cite journal |last1=Plantard |first1=Pierre |last2=Shum |first2=Hubert P. H. |last3=Pierres |first3=Anne-Sophie Le |last4=Multon |first4=Franck |title=Validation of an Ergonomic Assessment Method using Kinect Data in Real Workplace Conditions |journal=Applied Ergonomics |date=2017 |volume=65 |pages=562–569 |doi=10.1016/j.apergo.2016.10.015|pmid=27823772 |s2cid=13658487 |doi-access=free }}</ref>