Article: Recognition and analysis of key postures in sports basketball based on improved YOLOv3 algorithm Journal: International Journal of Reasoning-based Intelligent Systems (IJRIS) 2026 Vol.18 No.22 pp.19 - 31 Abstract: To address the issues of sensitivity of existing models to occlusion and motion blur, and to optimise the inference speed in multi-person or complex scenarios while reducing computational costs, this study proposes an improved single-stage object detection technique to enhance the accuracy of posture recognition and meet the daily training needs of basketball players. Experimental verification shows that the proposed model has achieved a mean average precision (mAP) improvement of 6.67% compared to the YOLOv3 baseline model. Additionally, compared to the previous version, the action recognition speed has increased by more than 50%, and the average accuracy has improved by 6.67%. This result provides reference value for computer vision-based motion recognition and target recognition algorithms, and lays a solid foundation for key posture recognition in basketball. Inderscience Publishers - linking academia, business and industry through research

Title: Recognition and analysis of key postures in sports basketball based on improved YOLOv3 algorithm

Authors: Yongquan Tang; Yuzhen Zhang

Addresses: School of Physical Education, Xinyang Normal University, Xinyang, 464000, China ' Anyang Vocational and Technical College, Anyang, 455000, China

Abstract: To address the issues of sensitivity of existing models to occlusion and motion blur, and to optimise the inference speed in multi-person or complex scenarios while reducing computational costs, this study proposes an improved single-stage object detection technique to enhance the accuracy of posture recognition and meet the daily training needs of basketball players. Experimental verification shows that the proposed model has achieved a mean average precision (mAP) improvement of 6.67% compared to the YOLOv3 baseline model. Additionally, compared to the previous version, the action recognition speed has increased by more than 50%, and the average accuracy has improved by 6.67%. This result provides reference value for computer vision-based motion recognition and target recognition algorithms, and lays a solid foundation for key posture recognition in basketball.

Keywords: YOLOv3 algorithm; loss function; model pruning; key posture recognition; basketball movement.

DOI: 10.1504/IJRIS.2026.155992

International Journal of Reasoning-based Intelligent Systems, 2026 Vol.18 No.22, pp.19 - 31

Received: 26 Jan 2026
Accepted: 24 Apr 2026
Published online: 28 Aug 2026 *

Title: Recognition and analysis of key postures in sports basketball based on improved YOLOv3 algorithm

Keep up-to-date