Comparison of SMAL and hSMAL Parametric Reconstruction of Horse Based on Point Cloud

With the increasing demand for animal breeding and the development of 3D reconstruction technology, an accurate and efficient 3D reconstruction method for horses is urgently needed. The aim of this paper is to investigate a parametric model-based 3D reconstruction method for horses. First, the 2D keypoints is detected using DeepLabCutand mapped to 3D keypoints based on camera parameters. The point cloud is then coarsely aligned with the model by minimizing the square error between the key points. Then, the 3D reconstruction of horses is transformed into a loss function minimization problem. The SMAL and hSMAL model parameters are gradually fitted to the point cloud data by iteratively optimising the shape, pose and translation parameters under the constraints of pose, keypoints and correspondence points to obtain the most appropriate mesh. The experimental results show that the hSMAL model can better fit the data in terms of posture and leg details than the SMAL model. In addition, the method shows greater robustness and adaptability in the face of missing data and noise interference in traditional reconstruction techniques. The results of the study provide an important technical support and data base for the automatic measurement of horse body size parameters.