Functional Combination And Migration With 3D Model Intelligent Construction

This paper deconstructed the problem of 3D model construction from the perspective of functions and proposed that artificial 3D models can be composed of structural and motion functions.Based on this theory,the main work and innovations are as follows.(1)As a precursor to the intelligent construction of the 3D model,a light field image based method for estimating the depth of the scenes is proposed.The method used projection error to self-supervise the network training for disparity estimation,which allows objects to be captured as dense 3D point clouds via a light-field camera.3D point cloud information of the object can then be used as a reference model and fed into the 3D model construction method.(2)Functional combination 3D model construction used deep learning to combine the functional structures of given 3D models and constructing a new 3D model with a reasonable structure.The approach proposes a generative autoencoder network that is trained using multiple 3D models of different categories and can reasonably assemble components from single or multi-category 3D models.The generative autoencoder constructs a new 3D model in different structural functions on cross-category models,or completing missing functions of the input 3D model on single-category objects.In addition,the experiments demonstrate the usability of the approach through a qualitative and quantitative comparison of structure combinations and shape complements.(3)Sample-based 3D model motion functions migration,can find a suitable motion structure from the model-set and migrated it into the input model.The approach allows the input model,which originally has only a static structure,to have a reasonable motion function.By analyzing the sub-structure of the input model and matching shape of the sub-structures with the sample,this approach can recommend a suitable sample.Based on the components association diagram of the input model and sample,the motion function of the sample can be migrated to the input model.Comparative experiments with similar methods can demonstrate the effectiveness of the approach.(4)Based on the two construction approaches described above,functional combination and migration with 3D model construction is proposed.And implemented a design recommendation system for multifunctional furniture that supports reconfiguration.The system used a given 3D furniture model as a reference to generate a multifunctional 3D model consisting of a set of plates of a specific shape.The multifunctional model can reconfigure its own structure through the motion of plates and achieve the same structure as the reference furniture.In summary,considering the structural and motion functions of 3D models,it first proposes 3D construction methods based on generative network assembly and sample-based motion migration,to demonstrate the feasibility of using functions to construct 3D models.The above ideas are then combined to present a design recommendation system for multifunctional furniture that supports reconfiguration.The system implements functional combination and migration with 3D model construction.