Title of the report: Reliable Body Modeling of Dynamic Objects in the Nature by Using Multi-sensor
Time: Thursday, 12-22-2011, 2:30pm
Place: Room 310, State Key Laboratory of CAD & CG, 3rd floor of Jiaoyi Building, Zhejiang University Yuquan Campus
Reporter: Dr. Li Guan (GE Global Research Center in America)
Host: Professor Geng Weidong
Abstract: Dynamic scene/event modeling with visual sensors (such as camera, infra-red camera, depth sensor, etc.) from multiple fixed positions is a problem of widespread interest in computer visual researches with rich actual applications, such as 3D TV, virtual reality, medical surgery, motion capture without marking, video games and security monitoring. However, most existing multi-angle systems are built in strictly controlled indoor environment and call for particular lighting condition and simple backgrounds. The application of that technology into outdoor environments will encounter many challenges, including varying light, shadows, reflection, mutual shading between modeled object and background objects, etc.. In this report, I will overcome above difficulties from two aspects -time and space- to decrease human manipulation and automate the modeling process. Two points are included: (1) a information fusion based general framework for computing 3D objects from the multi-angle system of heterogeneous sensor, which can also automatically reconstruct the static occluders with the silhouette information of foreground objects; (2) a algorithm for computing 3D “motion flow” of objects, where the bone structure of the modeled object is not necessarily known. “Motion flow” can help set up the consistency of natural motion time and initialize some complicated form research and motion analysis algorithm.