The Object Projection Feature Estimation Problem in Unsupervised Markerless 3D Motion Tracking

摘要: 3D motion tracking is a critical task in many computer vision applications. Existing techniques require either great amount of knowledge on the target object or specific hardware. These requirements discourage wide spread commercial applications based tracking. systems that no and run single low-budget camera estimations projection features (namely, area position). In this paper, we define feature estimation problem present novel system needs only requires camera, as installed most computers smartphones. Our estimates, real time, three-dimensional position non-modeled unmarked may be non-rigid, non-convex, partially occluded, self blurred, given it opaque, evenly colored, enough contrasting with background each frame. also able to determine relevant track screen. does not impose hard constraints, therefore allows market-wide implementation use