Research On 3D Human Body Measurement Method Based On Point Cloud Data

The size data of the human body has significant reference value to the garment industry.The accuracy of the data has a direct impact on the design and processing of clothing.The cost of traditional measurement methods(such as contact measurement)is relatively low.However,the measurement accuracy is largely affected by the experience of surveyors,which has some limitations.The non-contact measurement method utilizing structured light and laser scanning has high accuracy,but its high cost and operation difficulty limit its application in small and micro enterprises and individual users.As a kind of data with three-dimensional position information,point cloud can describe the shape of any object in three-dimensional space.By filtering,registration and reconstruction of point cloud,it can be applied to measure the size of human body.To reduce the measurement cost of the human body size and promote the accuracy of measurement data.This thesis adopts Kinect to collect point cloud data of the human body,and proposes a noval measurement method of the human body size based on these data,which can effectively reduce the measurement cost of the human body size while improving the accuracy of the measurement data.The details are as follows:(1)The traditional Iterative Closest Point(ICP)algorithm has high requirements on the position and pose of point clouds and poor applicability.We propose a noval an improved point cloud registration method,which takes the angle between the normal vectors of feature points as the judgment condition.Firstly,after filtering the point cloud,we calculate the normal vector of the feature point of the target point cloud and the point cloud to be registered.Then the initial pose of the point cloud is adjusted according to the angle between these normal vectors.The ICP method is exploited to further register the point cloud when the ideal pose state is obtained.The extensive experimental results on point cloud registration demonstrate that the proposed algorithm has high practicability and robustness,which ensures that the point cloud registration has high accuracy and reduces the registration time of the point cloud.(2)In order to resolve the problem of point cloud surface reconstruction,in this thesis,we calculate the normal line of the sample by constructing the covariance matrix of the point cloud and the normal vector by moving least square method.Then tangent plane projection and triangulation are utilized to reconstruct the surface.To verify the effectiveness of the proposed method,the abundant experiments have been carried out.The results suggest that the greedy projection triangulation algorithm has good performance in the reconstruction of details,reconstruction efficiency and precision,so that the reconstruction of the human body model can achieve the desired effect.(3)To measure the size of the human body,it is necessary to locate the feature points of the human body.Since human body keypoints depend on the curvature variation of the human body contour curve to a great extend,in this thesis,we propose the curvature variation of the feature interval fitting curve based method for extracting the human body feature points.First,the feature interval is located by exploiting the proportion relationship between the feature parts and the height of the human body.Then the least square method is used to fit the curve of the points in the feature interval.Finally,the human body keypoints are accurately located by calculating the curvature change of the fitting curve.The height,length and width of the human body are determined by calculating the shortest distance between the keypoints.The circumference data of the human body can be obtained by piecewise curve fitting of the point cloud at the height of the feature points and summing up the length of each curve obtained through integral operation.The trail results suggest that the proposed measurement method has high measurement accuracy,and its measurement error meets the expected standard.There are 29 figures,17 tables and 76 references.