Research On 3D Body Modeling Technology Based On Kinect

3D body model has been applied in clothing,medical diagnosis,game production and other fields,however,3D body scanning devices based on structure light and laser are not only bulky,difficult to operate,but also expensive,so they are difficult to be popularized in the market.Kinect is a depth camera,which can obtain the body's color image and depth data at the same time,then body's 3D point cloud can be constructed and body's model can be reconstructed according to the 3D point cloud.And Kinect is small in size,easy to operate and cheap,so now it has become a research focus of many scholars.The purpose of this paper is to study the 3D body modeling technology based on Kinect,and the main research contents are as follows:(1)In order to effectively extract body point cloud in point cloud data from Kinect,this paper proposes a method to make the target point cloud segmented.Firstly the method uses the Grab Cut algorithm in Open Source Computer Vision(Open CV)to segment color image,and then combines the segmented color image and corresponding depth information to generate point cloud.The generated point cloud only contains the target point cloud to be reconstructed and has no irrelevant redundant information,so the speed and precision of the subsequent model reconstruction can be improved.This method is simple,practical and pertinent,and has better performance compared with Euclidean cluster segmentation method.(2)Iterative Closest Point(ICP)algorithm is improved.Firstly the two pieces of point cloud are sampled by the voxel grid,then the Scale Invariant Feature Transform(SIFT)feature points of source point cloud are found and saved as point cloud,secondly the ICP algorithm is used to regist the saved point cloud and the target point cloud after voxel grid sampling.Moreover,the wrong matching point pairs are removed through Random Sample Consensus(RANSAC)algorithm,lastly the transformed source point cloud is saved and the process is repeated until the convergence condition is met.The validity of the method is proved by experiment.In addition,in order to solve the problem that the ICP algorithm can not register the two pieces of point cloud with poor initial position,a method of translating and rotating again before registration is proposed.The experimental results show that the method can register the two pieces of point cloud with poor initial position successfully.(3)Moving Least Squares(MLS)method is used to estimate the normal,to smoothand resample the point cloud data after registration,then Poisson surface reconstruction algorithm and greedy projection triangulation algorithm are introduced respectively,and modeling results of the two modeling algorithms are compared,lastly the body model reconstruction is realized by using the greedy projection triangulation algorithm.In this paper,the body point cloud data is collected by Kinect,and the point cloud segmentation,point cloud registration and point cloud reconstruction are studied respectively,which makes the 3D body modeling method based on Kinect more efficient and accurate.