Title: High resolution network combined with PnP algorithm for pose estimation of aerobics robot

Authors: Yan Liu; Yan Zhao; Bingyan Yu

Addresses: Public Courses Education Department, Anhui Business and Technology College, Hefei, 231131, China ' Public Courses Education Department, Anhui Business and Technology College, Hefei, 231131, China ' Public Courses Education Department, Anhui Business and Technology College, Hefei, 231131, China

Abstract: The traditional pose estimation method for aerobics robots has problems of low accuracy and computational efficiency. This study utilises high-resolution networks to extract local features and uses a transformer for multi-scale feature fusion to design a pose estimation method for aerobics robots. This method achieves an adaptive fusion of multi-scale features by constructing a transformer-enhanced HRNets feature extraction module, effectively capturing local and global features of robot joints. At the same time, a deformable attention mechanism is introduced to reduce the complexity of feature processing, and the EPnP algorithm with sparse control point constraints is used to establish the mapping relationship between 3D point clouds and 2D images. The precise solution of pose parameters is achieved through Levenberg-Marquardt optimisation. The proposed fitness robot pose estimation based on a high-resolution network and PnP algorithm can effectively improve the precision and efficiency of robot pose estimation, and reduce computational costs. This study is meaningful for the practical application of robot vision and attitude control.

Keywords: high-resolution network; PnP algorithm; transformer; aerobics robot; feature fusion.

DOI: 10.1504/IJSCC.2026.150317

International Journal of Systems, Control and Communications, 2026 Vol.17 No.1, pp.18 - 36

Received: 07 Feb 2025
Accepted: 26 Mar 2025
Published online: 09 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article