Human Body Acquisition System Based On Multi-view Camera

In recent years,with the continuous development and progress of 3D vision technology,human body motion capture has become a research hotspot in the field of computer vision,and it's also a very challenging research direction in the field of engineering.It has important applications in many fields,such as virtual reality,augmented reality,games,film production,etc.At present,the estimation of the pose of three-dimensional human body model mainly relies on some fixed large equipment,such as 3D scanning instrument,multi-camera system and so forth,or obtain the pose of human body by means of wearable devices and some clothes with special markers.Among them,the method with markers can obtain the human body model with high accuracy and good robustness,but such methods usually need to install different types of sensors or markers on the human body,which affects the movement of the human body partly and cannot reflect the most real movement state.In order to capture high-quality human motion and estimate human motion posture,this paper proposes a multi-view camera human motion capture system for markerless human motion capture.The main work is as follows:(1)Design and build a multi-view camera capture system,including camera selection,array construction,camera signal synchronization,image data acquisition and array calibration.The cameras in the camera system photograph of the moving human body synchronously,and obtain a series of continuous human motion image sequences that change with time.The 3D reconstruction algorithm is used to estimate the 3D structure of human body from the multi-view 2D images,and the 3D point cloud model of human body is obtained.(2)Estimating the pose of human body by OpenPose.Recovering the position of threedimensional key points in space according to the key points of human body in 2D image and triangulation method.This paper proposes a method for fitting SMPL model,three-dimensional key points,and three-dimensional point cloud of the human body.The method can solve the unnatural distortion at the joints of the human model due to the accumulation of errors in human motion capture,thereby estimating a more accurate human pose.(3)A dense human mesh template based on the SMPL model is proposed for human motion capture.By subdividing the original surface using the SMPL model,a dense human mesh model with more points and triangular patches is generated to describe more information,the template can be deformed towards the 3D point cloud model of the human body to recover the accurate human shape.Finally,the multi-view is used to shade the model and reconstruct the model's surface.Our motion capture method eliminates the constraint of the existing motion capture system on the catcher,and at the same time,the results of pose estimation are applied to the parametric model of the human body.Experiments show that our method improves the accuracy of human body attitude and makes the 3D human body model more realistic.