The Critical Techniques Of Virtual Fitting Room Using Skeleton Extraction

In this paper,we propose a new core algorithm for the virtual dressing room based on RGB-D camera.The algorithm can be roughly divided into two main modules,which are skeleton extraction and clothes deformation.First of all,the paper introduces some skeleton extraction and tracking techniques using depth map.Most of the state-of-art techniques give good results when the user faces the camera in their front face,but the performance degrades drastically when the user turns aside or turns around.Hence,the paper proposes a new skeleton extraction and tracking strategy to compromise the limitation of the state-of-art methods under some specific circumstances.This strategy can precisely measure the facing angle of the user's standing poses and effectively improve the performance when the user turns aside or turns around,compared with some skeleton detection algorithm(e.g.the embedded algorithm in Microsoft Kinect).Then,we refer to some common model deformation algorithms,including exterior cagedriven deformation strategy and skeleton-driven deformation strategy.At last,we build a complete system of virtual dressing room with the above two modules.To give users' better experience,we optimize the algorithm to meet the requirement of real-timing of the system.We test our system under some common poses which different people use when they fit clothes.The good performance demonstrates the effectiveness of our system.