Research On 3d Reconstruction Of Video Stream Based On Machine Learning

With the rapid development of graphics processing technology and display technology,human beings have higher and higher requirements for visual effects,which promotes the display equipment to achieve three-dimensional effect and bring real and vivid viewing experience to users.In recent years,the emergence of virtual reality market makes the research of stereo vision more and more important.It is difficult and expensive to collect stereoscopic video with camera,and the transmission bandwidth is large.However,for 3D scene reconstruction technology,only the color image and the corresponding depth image need to be transmitted in the process of video data transmission,which takes up less bandwidth.Therefore,3D scene reconstruction is the research direction in the field of machine vision.Reconstruction of 3D scene from video stream is an important research content in computer vision,image processing and pattern recognition.It is of great significance for the development of robot vision,UAV navigation,vehicle assisted driving and medical image analysis.In the aspect of depth map information extraction,this paper uses the graph cutting algorithm based on machine learning and combines the advantages of u-net and Mask r-cnn to extract more concerned foreground objects from images accurately.The corresponding background depth model is constructed according to the depth distribution structure of the scene,so as to obtain the background depth map.In order to better highlight the stereo effect of foreground objects,an additional depth value is assigned to each foreground object area based on the initial depth value to complete the foreground depth map.The final depth map can be obtained by combining the background and foreground depth map.Then,through the improved virtual viewpoint rendering technique,the image pairs for stereoscopic display are obtained.For the problem that there are holes in the reconstructed image,this paper summarizes the reasons for the formation of the holes,introduces the typical hole filling methods,and fills the holes by classification.In this paper,an improved neighborhood interpolation method is used to repair the small holes,the information in the reference image is used to fill the large holes.Through the depth gradient transition of foreground and background,the visual effect of the foreground and background boundary is enhanced.In this paper,the experimental analysis of 3D reconstruction technology of video stream based on machine learning is carried out through the specific data set,and the validity of the 3D reconstruction technology of video stream based on machine learning is verified.Compared with the traditional algorithm,the visual effect of stereo image is improved.Experimental results show that the resulting stereo image has good stereo effect.The experiment results show that the PSNR of image is 4.08%and 2.79%higher than that of criminisi algorithm and global variation algorithm,respectively.This algorithm also provides a feasible way for 3D reconstruction similar to the scene of vehicle road.