A dataset for human pose estimation using infrared imaging and low-cost reflective markers
- Ankith Motha
- Pranav Balaji R S
- Nanditha K
- Aravindhan B
- Saraswathi D
- Shyam Kumar S
- Davidson Jebaseelan
- Edward Jero S
2026-08-11
The infrared imaging pose estimation dataset (IRPose16) is a high-resolution human pose dataset that combines low-cost infrared hardware with reflective markers to enable reliable motion capture in low-visibility conditions. Unlike conventional datasets that rely on expensive multi-camera or RGB setups, it leverages an infrared modality for consistent performance under poor lighting, while focusing on body types representative of Indian populations that are often underrepresented in existing benchmarks. The dataset comprises over 100,000 frames from ten participants performing controlled full-body movements, each annotated with 16 anatomically defined keypoints. A fully automated detection and temporal linking pipeline generates dense annotations, followed by validation to ensure spatial consistency and temporal stability. The simplified marker-based setup reduces calibration complexity and supervision effort, making data collection scalable. IRPose16 is intended to serve as a reproducible resource for pose estimation, motion reconstruction, and applications requiring robust keypoint tracking in constrained or low-visibility environments.