Research On Several Key Techniques Of Data-Driven Virtual Fitting System

Supported by computer simulation technique,virtual fitting is capable of putting virtual clothes on the consumers.The visual presentation of virtual dressing allows consumers to enjoy the same experience as they are actually trying on clothes.In the numerous methods for querying consumer's information,depth data has been widely concerned and studied for it contains rich amount of infor-mation in the aspects of reconstruction and recognition.However,current depth data based virtual fitting systems commonly suffer several problems,such as the inadequate efficiency for body re-construction,the loss of body texture as well as the low accuracy of gesture recognition.So,how to incorporate the advantages of depth data into virtual fitting system while reduce consumer's waiting time is a very attractive question in the research community.In this thesis,we study the key techniques of depth data driven virtual fitting system.Our work focuses on sparse key points based parametric human body reconstruction,depth and colour camer-a calibration using hybrid parameters and combing KNN(K-NearestNeighbor)with LCSS(Longest Common Subsequence)for dynamic hand gesture recognition.The contributions in this thesis are listed as follows:[1]Parametric human body reconstruction based on sparse key points.We propose an automat-ic parametric human body reconstruction algorithm which can efficiently construct a model using a single Kinect sensor.A user needs to stand still in front of the sensor for a couple of seconds to measure the depth data.The user's body shape and pose will then be auto-matically constructed in several seconds.Our proposed scheme relies on sparse key points for the reconstruction.It employs regression to find the corresponding key points between the scanned depth data and some annotated training data.We design two kinds of feature descriptors as well as corresponding regression stages to make the regression robust and accurate.Our scheme follows with dense refinement where a pre-factorization method is ap-plied to improve the computational efficiency.Compared with other methods,our scheme achieves similar reconstruction accuracy but significantly reduces runtime.[2]Depth and colour camera calibration using hybrid parameters.We propose a linear optimiz-ing process for calibration,in which a matrix of hybrid parameters is introduced to linearize our optimisation.The hybrid parameters offer a transformation of a depth parametric space(depth camera image)to a colour parametric space(colour camera image)by combining the intrinsic parameters of depth camera and a rotation transformation from depth camera to colour camera.Both the rotation transformation and intrinsic parameters can be explic-itly calculated from our hybrid parameters with the help of a standard QR factorisation.We test our algorithm with both synthesised data and real-world data where ground-truth depth information is captured by Microsoft Kinect.The experiments show that our approach can provide comparable accuracy of calibration with the state-of-art algorithms while taking much less computation time(1/50 of Herrera's method and 1/10 of Raposo's)due to the advantage of using hybrid parameters.[3]Combing KNN with LCSS for dynamic hand gesture recognition.We propose a dynamic hand gesture recognition system based on Nguyen-Dinh's longest common subsequence(LCSS)based method,and improve the recognition accuracy from two aspects.First,inorder to pro-vide discrete features which characterize gesture movement for LCSS,this paper presents a direction discretization method to encode gesture features which can show the differences of each gesture.Second,taking into account different people have different speeds,trajectories and spatial positions to perform the same gesture,this paper combines K nearest neighbors(KNN)with the LCSS to solve the different habits of users.The experiment results show that the recognition accuracy of dynamic hand gesture is benefited with our combining KNN with LCSS method.(F1-score).