Research On Image-based 3D Reconstruction Algorithm Of Giant Panda

As a popular research direction in the field of computer vision,3D reconstruction has been widely used in fields such as autonomous drive,digital archaeology,and medical 3D images.As a national first-level protected animal in China,the giant panda can truly and completely reflect the spatial shape structure of the giant panda through the three-dimensional reconstruction of the giant panda,and disseminate and display the giant panda model in the most intuitive form.The panda body measurement method is changed from the traditional contact type to the non-contact body measurement,which greatly reduces the cost and saves resources while greatly improving the efficiency;analysis of the body measurement data obtained through the three-dimensional reconstruction of the panda can be timely Understand the health status of giant pandas and provide better protection for the growth and breeding of giant pandas.Because pandas are non-rigid objects and have the characteristics of easy deformation,this makes the 3D reconstruction of pandas very difficult,and the related research data is difficult to obtain,so the research on 3D reconstruction of pandas at home and abroad is still a blank field.Although the three-dimensional reconstruction of animals is still in its infancy,the three-dimensional reconstruction technology based on the Skinned Multi-Animal Linear Model(SMAL)model is already the mainstream method for completing the three-dimensional reconstruction of tetrapods in the world.It has high accuracy,good effect and simple reconstruction process.Therefore,through the analysis of the SMAL model,we will apply the SMAL model to the research of 3D reconstruction of giant pandas,and achieves a good 3D reconstruction effect of giant pandas.The main research content of this paper is divided into the following parts:(1)Research on SMAL model of giant panda: SMAL model is a parametric animal model based on animal skeleton and joint modeling.We analyze the difference between the giant panda's skeletal structure and the quadruped in the SMAL model,and combine the giant panda's skeletal structure characteristics to construct the giant panda SMAL model.Further,the principal component analysis is used to estimate the shape basis of the panda SMAL model,and the pose basis in the panda SMAL model is obtained through the skeletal motion representation,thereby achieving the parametric expression of the panda SMAL model.Finally,the experiment of controlling the deformation of the three-dimensional panda model through shape parameters and posture parameters verifies the feasibility of the parameterized panda SMAL model to realize the three-dimensional reconstruction of the giant panda.(2)Research on image-based 3D reconstruction algorithm of giant panda: design an energy optimization algorithm to fit the giant panda 2D image to the giant panda SMAL model to realize the restoration of the giant panda 3D model.In this paper,the panda contour segmentation map and key points are manually extracted from the input two-dimensional image;by constructing the target energy function and minimizing it,the panda SMAL model can fit the image as much as possible,thereby estimating the panda SMAL model's The parameters such as shape and posture will eventually realize the three-dimensional reconstruction of the giant panda.(3)Research on 3D reconstruction algorithm of giant pandas based on deep learning: A residual network model based on attention mechanism is proposed for image feature extraction.The entire network model directly conducts shape parameters and pose parameters in the panda SMAL model from 2D images.Estimate and predict,and then realize the 3D reconstruction of the giant panda,without the need for image segmentation and manual selection of key points.The experimental results show that,compared with the non-linear optimization fitting algorithm for manually extracting key points of the image,this algorithm can automatically extract the relevant parameters of the giant panda SMAL model while ensuring the accuracy of the three-dimensional reconstruction of the giant panda.Compared with the network model that does not use the attention mechanism,this algorithm can effectively improve the accuracy of the 3D reconstruction of the giant panda.At the same time,we also use the reconstructed 3D model of the giant panda to complete the body measurement experiment of the giant panda.