I. Introduction

In recent years, the analysis of human movement dynamics has gained increasing prominence across multiple domains, including sports medicine, rehabilitation, ergonomics, and the design of biomechanical devices [1]–[3]. In fact, accurate measurement of human posture plays a crucial role in enhancing athletic performance, supporting clinical diagnosis, and preventing injuries [4], [5]. In biomechanics and sports engineering, 3D human pose analysis enables the reconstruction of joint geometry and provides information on body kinematics during various activities, such as walking, running, and cycling [6]. Among these, cycling presents unique challenges, as it involves not only complex movements but also continuous human-machine interaction, which introduces additional sources of variability and uncertainty in the estimation of 3D joint positions. Marker-based motion capture techniques, such as optoelectronic systems and inertial measurement units (IMUs) [7], are commonly employed for 3D Human Pose Estimation (HPE) and are considered as the standard in terms of metrological performance. Nevertheless, these systems present significant practical limitations. They are primarily intended for use in laboratory settings and are difficult to implement in real-world or outdoor environments. Furthermore, the application of external markers or sensors may constrain the athlete's natural movements or be occluded by equipment such as the bicycle. Furthermore, the accuracy of such systems is highly sensitive to marker placement, which can introduce systematic measurement errors. In addition, external sensors or markers may shift during the course of testing due to factors such as perspiration and compromise the consistency and accuracy of the recorded measurements [8].