Human body volume (BV) is a useful biometricfeature for human identification and an important medicalindicator for monitoring body health. Traditional BV estimationtechniques such as underwater weighing and air displacementdemand a lot of equipment and are difficult to be performedunder some circumstances, for example, in clinical environmentswhen dealing with bedridden patients. In this contribution,a novel vision-based method dubbed Point2PartVolume basedon deep learning is proposed to rapidly and accurately predictthe part-aware BVs from a single-depth image of the dressedbody. First, a novel multitask neural network is proposed forjointly completing the partial body point clouds, predictingthe body shape under clothing and semantically segmentingthe reconstructed body into parts. Next, the estimated bodysegments are fed into the proposed volume regression networkto estimate the partial volumes. A simple yet efficient twostep training strategy is proposed for improving the accuracyof volume prediction regressed from point clouds. Comparedto existing methods, the proposed method addresses severalmajor challenges in vision-based human BV estimation, includingshape completion, pose estimation, body shape estimation underclothing, body segmentation, and volume regression from pointclouds. Experimental results on both the synthetic data andpublic real-world data show that our method achieved average90% volume prediction accuracy and outperformed the relevantstate-of-the-art.
Hu, P, Dai, X, Zhao, R & Munteanu, A 2023, 'Point2PartVolume: Human Body Volume Estimation from A Single Depth Image', IEEE Transactions on Instrumentation and measurement, vol. 72, 5502812. https://doi.org/10.1109/TIM.2023.3284948
Hu, P., Dai, X., Zhao, R., & Munteanu, A. (2023). Point2PartVolume: Human Body Volume Estimation from A Single Depth Image. IEEE Transactions on Instrumentation and measurement, 72, Article 5502812. https://doi.org/10.1109/TIM.2023.3284948
@article{5ac82c31cdaa4177be82e9545da4daf7,
title = "Point2PartVolume: Human Body Volume Estimation from A Single Depth Image",
abstract = "Human body volume (BV) is a useful biometricfeature for human identification and an important medicalindicator for monitoring body health. Traditional BV estimationtechniques such as underwater weighing and air displacementdemand a lot of equipment and are difficult to be performedunder some circumstances, for example, in clinical environmentswhen dealing with bedridden patients. In this contribution,a novel vision-based method dubbed Point2PartVolume basedon deep learning is proposed to rapidly and accurately predictthe part-aware BVs from a single-depth image of the dressedbody. First, a novel multitask neural network is proposed forjointly completing the partial body point clouds, predictingthe body shape under clothing and semantically segmentingthe reconstructed body into parts. Next, the estimated bodysegments are fed into the proposed volume regression networkto estimate the partial volumes. A simple yet efficient twostep training strategy is proposed for improving the accuracyof volume prediction regressed from point clouds. Comparedto existing methods, the proposed method addresses severalmajor challenges in vision-based human BV estimation, includingshape completion, pose estimation, body shape estimation underclothing, body segmentation, and volume regression from pointclouds. Experimental results on both the synthetic data andpublic real-world data show that our method achieved average90% volume prediction accuracy and outperformed the relevantstate-of-the-art.",
author = "Pengpeng Hu and Xinxin Dai and Ran Zhao and Adrian Munteanu",
note = "Publisher Copyright: {\textcopyright} 1963-2012 IEEE.",
year = "2023",
month = jun,
day = "9",
doi = "10.1109/TIM.2023.3284948",
language = "English",
volume = "72",
journal = "IEEE Transactions on Instrumentation and measurement",
issn = "0018-9456",
publisher = "Institute of Electrical and Electronics Engineers",
}