| With the rapid development of computer science and technology,garment industry is gradually changing to the efficient and user-centered model.Virtual fitting obtains the widespread attention.How to use computer technology to measure parameters of human body correctly is a key problem in virtual fitting area.It is also an important research topic in machine learning and computer vision.In this paper,the gradient boosting regression tree is used to explore and study the problem of remote body parameter estimation.We construct features from 2D images.Through interactive Bezier curve fitting,we get waist,bust,hip and other circumference.After eliminating some unimportant features,we can increase the accuracy of prediction to 78.3%.Finally,we demonstrate the effectiveness of our model with many experiments and comparisons.In addition,our paper presents an efficient and robust remote body parameter estimation framework based on 3D reconstruction.It supports different types of input data,like a reconstructed human body model or multi-view image sequences.We can use the geodesic distance to measure the fully reconstructed model directly,which can be reconstructed through an RGB-D camera.As for the input multi-view image sequences,firstly,we obtain the camera pose and the depth of image sequence through camera tracking and depth recovery techniques.At the same time,we eliminate redundant 3D points through sampling and multi-view fusion.Then the transformation matrix is solved by feature matching.As for some special cases,we use manual annotation to replace traditional SIFT features,which makes our algorithm more robust.Finally we use the iterative closest point algorithm to refine our model and take the ratio of the actual height to the measured height in the model as the scaling ratio,then we can use the geodesic distance to measure the length of any part of the human body,and further improve the accuracy of prediction by combining the GBRT algorithm and 3D reconstruction. |
PDF Full Text Request