Fast Reconstruction Of 3D Human Body Sitting Posture Based On KinectFusion Technology

Three-dimensional geometric data of human body sitting posture plays an important role in the development of riding tool and human body model.For example,the design of car seats is based on the datum of human contour figure and the development of anthropomorphic test device and dummy FE model which are often used in the automobile safe field needs more detailed human sitting data.At present,the existing human sitting data is mainly from manual measurement and traditional three-dimensional scanning device.Although the manual measurement is simple,convenien and low cost,the workload is large and it's prone to error.Compared with manual measurement,the traditional three-dimensional scanning device has high measuring speed and high precision,but it is bulky,expensive and complicated to operate.Therefore,it is of great value to study the three-dimensional human sitting posture measurement system with small volume,low cost and high precision.In view of the above situation,this paper,using Microsoft's Kinect depth camera,designs and implements a three-dimensional human body sitting reconstruction system based on Microsoft KinectFusion technology.Some algorithm which were used for removing background point cloud,smoothing and filling hole were developed.The difficulties like 3D reconstruction model has holes and was error-large were solved.The main research work of this paper is as follows:1.The influence of the human sitting posture on results of 3D reconstruction model was explored,and the best sitting posture was determined.The human body sitting support platform was designed for solving occlusion problem.2.By using the Microsoft KinectFusion and custom point cloud processing algorithm,the collection and processing of the human point cloud data was achieved.By comparing the acquisition methods of the two kinds of point cloud,KinectFusion was determined to obtain point cloud data.Point cloud was simplified and the background point cloud was removed through the straight through filter;An algorithm,combining automatic segmentation of human body with the hole filling based on least squares,was presented to fill holes.3.Influencing factors of Kinect scanning accuracy was optimally analyzed.According to the results of orthogonal experiment,the influence of the three factors Kinect height,Kinect from the circular turntable center distance,turntable speed on the human sitting posture scanning accuracy was discussed.And the results go through the approximate model fitting and parameter optimization,in which process,the best Kinect placement and turntable speed was got.4.The validity of three dimensional human body sitting reconstruction system was tested by the volunteers' three-dimensional scanning experiment.In this paper,the measurement parameters of the surface feature size of the human body were determined,and the measured body size parameters of the human sitting posture were compared with the real data of the human body to evaluate the accuracy of the reconstructed results of the three-dimensional human body.The research results show that the three-dimensional human body sitting reconstruction system based on Microsoft's KinectFusion technology can be used to reconstruct 3D human body,and through its designed specific body sitting support platform,the problem of occlusion in the process of sitting position scanning can be effectively solved;Through background removal algorithm,hole filling algorithm,smoothing algorithm and surface reconstruction algorithm,the complete three-dimensional body sitting model can be obtained.And the average error of the 3D model of human body sitting posture is 0.9cm,with reconstruction precision being high.Besides,the system has the advantages of low cost,simple operation and so on.The establishment of 3D geometric reconstruction method for human sitting posture will provide support for human body posture data acquisition.