Research Of Human Body Model Reconstruction Based On Depth Camera

Virtual reality and augmented reality are one of the hottest computer technologies at present.The reconstruction of 3D human body model is an important research direction of virtual reality and virtual reality.It can be used in a wide range of virtual reality games,the production of movie special effects,the virtual test of the reality class,and 3D printing.In the past,3D human models were usually acquired by large scale laser equipment or stereo imaging technology.This paper use Kinect to get images of the human body from different angles.Using ORB combined with feature point pair filtering algorithm to find physical points in different dimensions in the data,then using PnP to solve different dimensions of camera position and pose.Then,using local optimization registration algorithm,global optimization registration algorithm,and non rigid registration algorithm to realize the registration of human point clouds.Finally,using point cloud reconstruction algorithm to reconstruct human models.This method solves the expensive cost of large equipment purchase,and reduces the complexity of stereo vision algorithm in ensuring the speed and quality of human reconstruction.This paper improves the ORB feature extraction algorithm,and proposes an algorithm for sifting the erroneous feature point pairs.After using the Kinect depth camera to obtain the data,in order to achieve real-time data processing,the ORB algorithm is used to extract the panoramic RGB image to get the projection points of the object in different positions.In view of the shortcoming of ORB feature extraction algorithm,the right matching points are selected according to the direction of the main motion.For all data alignment,this paper proposes a key frame selection algorithm for human point cloud registration to solve the data redundancy.To solve the problem of point cloud registration,the local registration optimization method is adopted to reduce the error.At the same time,the global optimization registration,loop detection,non rigid registration are used to solve the error accumulation,realize the closure of the loop,solve the problem of human body movement,and improve the registration accuracy.In this paper,the human body model is constructed by using Kinect to collect data from each direction.Through the experiment,this paper can get a clear human body model quickly.