3D Human Pose Estimation from Monocular Video Sequences in Underwater Scenarios
Shuwen Liang, Hailong Liu, Ping Liu, Dong Zhang, Rong Yu, Xiaowei Zhou, Zhize Zhou · 2026
Read 3D Human Pose Estimation from Monocular Video Sequences in Underwater Scenarios by Shuwen Liang, Hailong Liu, Ping Liu, Dong Zhang, Rong Yu, Xiaowei Zhou, Zhize Zhou (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13469234.txt.
Description
Shuwen Liang and colleagues estimate three-dimensional human pose from a single underwater camera in 2026, a setting where refraction bends light, bodies occlude themselves, and image quality collapses. Their two-stage method first runs a part-attention regressor that fits a parametric body model and is robust to occlusion. A detector pipeline alongside finds the person and regresses 2D keypoints even in murky frames.
An optimization stage then applies an underwater variational autoencoder to correct pose parameters that refraction and occlusion corrupted. The autoencoder is a learned model of how human bodies plausibly move underwater. Constraints keep the joints aligned with the image evidence while the prior pulls impossible poses back to legal ones. Underwater pose data serve swim training, diving safety, and marine robotics. A monocular method avoids the cost of multi-camera tanks. That is why the authors built the prior into the estimator itself.
Catalog record
| CREATOR | Shuwen Liang, Hailong Liu, Ping Liu, Dong Zhang, Rong Yu, Xiaowei Zhou, Zhize Zhou |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |