Research On 3D Human Reconstruction Technology Based On Single Depth Map

Three dimensional human reconstruction means that the real human form is input into the computer through computer vision,and its application area is quite extensive.Nowadays,mainstream 3D human modelling uses large-scale 3D scanning instruments for human body scan reconstruction.However,it requires the user to stand in front of the device for a long period of time,and it cannot be used in any scene,and the installation and use of the device is extremely complicated.Although the modeling accuracy is very high,the computation time is long and the modeling efficiency is low.The reconstructed model will have holes due to the self shielding of the scanned object.The vertex of the model does not have a good topological relationship.In order to overcome the shortcomings of traditional human reconstruction methods,we propose a human body modeling system using a single Kinect depth image.The system uses Kinect to instantly collect a single frame frontal depth data,and does not need to scan the user for a long time.As the single front depth data is missing back data,this paper uses a parametric variable model with prior knowledge to fit the human depth data,and estimates the back of the human body to obtain a human reconstruction model that approximates the real scan data.The work of this article mainly includes the following sections:1.Due to the inconsistency between the height of the variable model and the height of the collected human point cloud,this paper proposes a method based on the height of the bounded bounding box of the object to unify the variable model and the height of the human point cloud.2.Since the parameterized variable model fits the human point cloud data needs to establish an accurate correspondence relationship,this paper proposes an initial registration method based on the directed bounding box of the object to roughly register the variable model and the human body point cloud,and then adopts the closest point iterative method with normal constraint is used to establish the exact correspondence.3.The optimized skeletal data is used to roughly fit parametric variable models and human point cloud data poses,and an improved linear mixed skin technology is used to accurately fit the pose of the parameterized variable models and human point cloud data.The system makes full use of the advantages of low cost,convenience,and shortcut of Kinect.It combines the advantages of variable models to overcome the shortcomings of low precision of Kinect data acquisition and holes in traditional scan reconstruction.From the system reconstruction results,it can be shown that using a single human body depth data for human reconstruction can generate a better grid model.