The Research And Realization Of 3D Human Reconstruction Based On Kinect

With the rapid development of computer vision,3D human body reconstruction technology has been widely used in all walks of life,especially in film and television animation,human-computer interaction,virtual fitting and augmented reality has a greater application requirements.In this paper,we choose a 3D human body reconstruction method based on Kinect.Compared with other methods,this method has the advantages of low cost,easy operation and the data are collected not affected by environment can quickly obtain the depth information of objects,is an effective method,this paper mainly carried out the following points:Firstly,the calibration of the Kinect are studied and improved.For the color and depth camera in the Kinect,the depth and color information can not be aligned precisely due to the assembly and lens distortion.This paper presents an improved Harris algorithm to calibrate Kinect,the algorithm is mainly by optimizing the Harris corner response function to make the calibrated Kinect reduce the projection error,and to improve the calibration accuracy,further gives the relative conversion between the two cameras,it is has made a very good foundation for the point clouds data.Secondly,in order to solve the noise problem in point clouds data,this paper presents a denoising algorithm based on point clouds noise classification.The noise is classified in this algorithm,the large-scale noise is removed combines the radius filtering and statistical filtering.Fast bilateral filtering is used to smooth the small-scale noise.The proposed algorithm completes the denoising and smoothing for the point clouds data.Compared with the traditional bilateral filtering algorithm,the algorithm greatly reduces the time complexity and the denoising effect is also better.Then,the registration of point clouds data is studied,and the experimental results are given.The algorithm is mainly to convert the point clouds data from different perspectives into the unified coordinate process,which is mainly divided into two stages,namely coarse registration and fine registration.The coarse registration uses the sample consensus initial alignment algorithm based on FPFH to obtain the initial position between the point clouds pairs,and then uses the Kd-tree based ICP fine registration method to complete the final registration,make the error is minimized.At last,the method based on greedy projection triangulation and Poisson surface reconstruction method is given,and the experimental results are given respectively,then the surface reconstruction of the point clouds model is completed.The experimental results show that the proposed 3D human body reconstruction method in this paper can complete the reconstruction of the human body model well and the reconstruction process is simple.The experimental environment can reconstruct the human body quickly without complex and expensive equipment.And the reconstructed human body accuracy is higher,but the details of the reconstruction effect is not good enough,it can meet the basic requirements for the less demanding application.