3D Animation Generation Method Based on Pose Estimation Algorithm

The current 3D human animation generation process requires a significant amount of manpower and material resources to adjust the model accordingly. In order to make 3D animation generation more efficient and reliable, a multi task and multi-level neural network is used to complete human pose estimation, and the instance calculation is divided into two working threads to optimize the bandwidth consumption of the animation generation system, effectively improving the speed of animation generation. The test results show that the training time of the 3D pose estimation model based on deep learning is between 39ms and 42ms, while the average training time of the pose estimation model based on adaptive fusion units is 43.92ms. The proposed 3D animation generation system has improved accuracy by 30.20% compared to animation generation systems based on prior models. It can be seen that using this improved 3D animation generation method can achieve real-time and accurate 3D human animation generation, which has a positive promoting effect on the development of virtual creation, art exhibitions, augmented reality and other fields.