Indoor Moving Human Detection And Dynamic 3D Reconstruction

With the development of artificial intelligence,computer vision technology has been widely used,and has become a hot topic of research.Moving human detection and dynamic human body 3D reconstruction are both hot and difficult topics in the field of computer vision.The research is studied in indoor environment in this thesis.Depth camera is used to collect data.On the basis of moving human body detection,the dynamic 3D reconstruction of human body is carried out.Firstly,the research background and related basic theories are introduced.From the aspect of imaging principle of deep camera,the cause of noise in depth image is analyzed.Aiming at the two kinds of noise contained in depth image: black holes and edge burr,the black holes is filled and the image is smoothed to remove the noise contained in depth image.Secondly,foreground-background segmentation is applied to the image after removing noise.Based on the foreground depth image,a novel real-time moving human detection algorithm is proposed.In the environment of multi-person,multi-pose and occlusion,the human body in the scene is detected and labeled by synthesizing the edge length and physical height of the foreground objects.Experimental results show that compared with the existing similar algorithms,the proposed human detection algorithm in this thesis has high detection accuracy and fast processing speed.On the basis of human detection,a dynamic human body 3D reconstruction algorithm using Laplacian deformation method for fine reconstruction of large deformation region is proposed.The rough reconstruction model is obtained by volume warp field,and the large deformation region is optimized by Laplacian deformation to achieve a fine human body reconstruction model.The proposed algorithm does not need a priori template,and can reconstruct a more realistic human model in the case of large deformation.Finally,this thesis summarizes the work of this thesis,analyses the advantages and disadvantages of proposed algorithms,and prospects the future research work.