Machine vision 3D modeling technology has received more and more attention due to its great commercial application value, and has become a fundamental technology in multiple fields. The article proposes a multi-view video 3D target modeling method combining SFM and NeRF network. The growth of seedling plants is taken as the research object, and the multi-view video of seedling plants is obtained by recording video. Python development platform was used to obtain the estimation of image position through COLMAP software, on the basis of which the image position was inputted into NeRF network to realize the 3D target modeling of multi-view video, and the similarity matrix and KNN algorithm were used to cloud the 3D model, and statistical filtering was introduced to remove the outliers of the point cloud. The combination of SFM and NeRF can significantly enhance the accuracy of multi-view video 3D target modeling, and the overall reconstruction efficiency and quality are high. Therefore, actively exploring the application of deep learning techniques in multi-view 3D target modeling can further promote the development of 3D target modeling technology.
