As the use of 3D modeling becomes more widespread, animation 3D techniques are facing more challenges. Neural radiation field provides a new idea to solve this problem by virtue of its ability to reconstruct a realistic 3D scene from sparse 2D images. In this paper, we use the depth camera (i.e., RGBD camera) to obtain the image depth information and generate the 3D point cloud data, combined with the greedy projection triangulation algorithm to reduce the 2D triangular mesh mapping to 3D space to form a 3D space triangular topology network structure, and obtain the 3D surface model of the object. Improve the neural radiation field for animation scene viewpoint synthesis, and design the neural radiation field framework based on space distortion. Comparison tests are carried out to analyze the reconstruction performance and rendering effect of the two algorithms. The greedy projection triangulation reconstruction algorithm takes only 12.043s to process the instance horse, and the maximum deviation distance, average deviation distance, standard deviation and root mean square error do not exceed 0.1 mm. The viewpoint synthesis method based on the improved neuroradiometric field shows certain algorithmic advantages in the viewpoint synthesis of animated scenes.
