Research On 3D Human Body Model Reconstruction Based On Kinect In Virtual Dressing Room

The virtual fitting environment is a technology which alows users view their fitting results on a display screen without needing to actually change clothes on their bodies.In other words,with the help of electronic devices,people can easily view and compare different fitting effects on their display screens.Along with development of the humancomputer interaction and the maturity of somatosensory technology,virtual fitting technology has also received a great attention,while the human body 3D model reconstruction technology will be the core content of virtual fitting technology.The reconstruction of 3D human body model is to establish a mathematical model for the human body,and the model is suitable for computer representation and processing.It is a virtual reality technology that expresses the objective world through computer.This research focuses on developing 3D human body models for virtual fitting environment.The technology basis applied here is the Kinect equipment,which makes it possible to develop 3D human body models at a high precision and low costs,and this will provide the theoretical basis for the study of the virtual fitting room system.This article will introduce how can the color data in human body image can be obtained by using a second-generation Kinect depth camera.Meanwhile,it also reveals how to reconstruct a dynamic 3D model of the human body through the 3D dynamic fusion technology.Meanwhile,all reconstructed models will also be optimized while they are output from the system.Moreover,without the prejudice to accuracy and completeness,in order to make the output model more suitable for the virtual fitting environment,methods such as boundary extraction will be applied to reconstruct the 3D human body image area.The key advantage of using DynamicFusion real-time dynamic 3D fusion technology is acquiring dynamic 3D human body data while the Kinect camera is fixed in a certain place.To be specific,the principle of dynamically reconstruct the 3D model of the human body is to set up key frame models and allow the obtained 3D data to dynamically match the key frame models.This technology has not only simplifies the human body 3D data acquisition process and alows the subject to perform arbitrary actions during the model reconstruction process,but also more suitable for the real fitting environment.The main work done in this paper is as follows.1.Build a second generation Kinect equipment acquisition environment and data processing platform in order to achieve the immediacy and accuracy requirements for human body 3D model in the virtual fitting room environment.Besides,perform bilateral filtering,coordinate alignment,etc.on the human body image color data and depth data acquired by the second Kinect equipment.2.Perform further pre-processing on the pre-processed standard data in a virtual fitting room environment.What's more,limit the reconstructed area as the human image area by applying contour extraction method,so that the amount of data processed in the later processing period will be reduced and the output model will be more suitable for the virtual fitting environment.3.Realize DynamicFusion real-time dynamic 3D fusion technology,namely,a realtime dynamic 3D fusion of preprocessed data.The three-dimensional model reconstruction of the human body will be realized by establishing a key frame model,solving the body warping field and real-time three-dimensional data fusion according to the body warping field.4.Experiment with four people and rebuild their 3D body models with current platform.Firstly,four body parameters that have a large impact on the virtual fitting are selected.Afterwards,the standard human body static measurement data is used to reconstruct the model five times for each object.The final step is to weight and average the results and compared them with the actual manual measurement data.Through this experiment,the model parameter error and system error can be accurately obtained.5.According to the comparison between the 8 sets of experimental data and the actual manual measurement data,it is found that the average error of each set of data is 2%-6%.Therefore,it turns out that the average error fluctuation of the 8 sets of data is small while the overall system error is also stable.Accordingly,it has proved that the DynamicFusion algorithm can provide suitable 3D human body model for the virtual fitting environment.