Key points are not available for this paper at this time.
3D human pose estimation is pivotal in the domains of computer vision and kinematic analysis. However, the majority of existing 3D datasets focus on indoor environments, resulting in a significant gap in extensive research on outdoor and specific athletic settings. This gap is notably evident in activities like outdoor running, where there is a stark lack of adequate data for reliable pose estimation. Contemporary motion capture systems face challenges in capturing precise ground truth data in outdoor settings, highlighting the necessity for alternative methods to address this shortfall. In response, our study introduces a dataset created using the Unreal Engine (UE) simulation environment, meticulously designed for assessing poses in outdoor multi-person running scenarios. This dataset comprises a diverse range of human models, running speeds, and gait patterns, striving to mirror a wide array of outdoor running conditions. To mimic real-world intricacies, the dataset integrates factors such as occlusion and varying lighting conditions. We provide high-fidelity ground truth data, free from temporal delays, to enhance the evaluation of algorithmic performance. Through thorough validation and experimental scrutiny of our dataset, its efficacy and utility have been demonstrated. This dataset stands as a critical resource for studies in outdoor multi-person running pose estimation, adeptly filling the void in data for outdoor athletic scenarios. This research substantially aids in furthering the exploration and development in the field of 3D human pose estimation.
Tong et al. (Fri,) studied this question.