The pose estimation model plays a pivotal role in representing the human body for action recognition systems. Various software and hardware solutions are available for building pose estimation models, with well-known examples including Microsoft Kinect, Intel Realsense with the Cubemos SDK, Nuitrack, BlazePose (for 3D models), as well as AlphaPos, PoseNet, OpenPose, and YOLOv11 (for 2D models). However, many action recognition systems are developed for specific models, resulting in dependencies on certain hardware and software. These dependencies complicate the training of neural networks across multiple datasets, limiting the scalability of human activity monitoring applications. To address these challenges, we propose a standardized pose estimation model along with methods for transforming between different model formats. This approach aims to overcome compatibility and interoperability issues in heterogeneous computing environments. Given that pose models can be either 2D or 3D, their integration into action recognition systems poses additional complexity. To simplify this, we introduce activity map tuning tailored to 2D pose models. Our experimental studies validate the effectiveness of these transformation methods and activity map tuning. The results show that the standardized model preserves accuracy while enabling the development of more adaptable and efficient human activity monitoring systems. This flexibility enhances the potential for broader applications and improved real-world performance.