Monocular 3D Human Modeling for Pediatric Anthropometric Estimation ■
Subject ■
Accurate measurements of body weight and height are essential for assessing growth,
nutritional status, and clinical decision-making in pediatric populations. In particular, body weight is
crucial for the correct dosing of medications and fluid administration in children. However, direct
measurement is often not possible, for instance in emergency situations or with neurological impaired
children who are not able to stand. In these situations, clinicians often estimate weight using
regression formulas based on surrogate anthropometric measures, such as wrist measurements.
These formulas are only approximations and are inaccurate.
Recent advances in computer vision have opened new possibilities for estimating human body shape
from ordinary RGB images. In particular, parametric human body models such as SMPL [1] represent
the human body as a deformable 3D mesh whose shape and pose can be adjusted to match an
observed person in an image. Fitting such models to single-camera images of children could enable
the extraction of body dimensions and shape descriptors that are informative for estimating body
weight and body composition.
In collaboration with UZ Brussel, a dataset of children has already been collected, including RGB
images acquired in multiple standardized poses against a clean background, together with groundtruth
body weight and a large set of anthropometric measurements.
nutritional status, and clinical decision-making in pediatric populations. In particular, body weight is
crucial for the correct dosing of medications and fluid administration in children. However, direct
measurement is often not possible, for instance in emergency situations or with neurological impaired
children who are not able to stand. In these situations, clinicians often estimate weight using
regression formulas based on surrogate anthropometric measures, such as wrist measurements.
These formulas are only approximations and are inaccurate.
Recent advances in computer vision have opened new possibilities for estimating human body shape
from ordinary RGB images. In particular, parametric human body models such as SMPL [1] represent
the human body as a deformable 3D mesh whose shape and pose can be adjusted to match an
observed person in an image. Fitting such models to single-camera images of children could enable
the extraction of body dimensions and shape descriptors that are informative for estimating body
weight and body composition.
In collaboration with UZ Brussel, a dataset of children has already been collected, including RGB
images acquired in multiple standardized poses against a clean background, together with groundtruth
body weight and a large set of anthropometric measurements.
Kind of work ■
Objective:
The objective of this thesis is to develop and evaluate a monocular vision-based pipeline for estimating
body weight and anthropometric parameters of children using 3D human body models fitted to RGB
images.
Description of work:
• Literature review on monocular 3D human reconstruction and anthropometric estimation
• Implementation of a pipeline to fit a 3D body model to pediatric images
• Extraction of body shape parameters and anthropometric features from the fitted model
The objective of this thesis is to develop and evaluate a monocular vision-based pipeline for estimating
body weight and anthropometric parameters of children using 3D human body models fitted to RGB
images.
Description of work:
• Literature review on monocular 3D human reconstruction and anthropometric estimation
• Implementation of a pipeline to fit a 3D body model to pediatric images
• Extraction of body shape parameters and anthropometric features from the fitted model
Expected Student Profile ■
(Mandatory) qualifications:
• Following a MSc in a field related to one or more of the following: Computer
Science, Biomedical Engineering, Applied Computer Science - Digital Health.
• Strong programming skills (Python).
• Ability to write scientific reports and communicate research results in English.
• Following a MSc in a field related to one or more of the following: Computer
Science, Biomedical Engineering, Applied Computer Science - Digital Health.
• Strong programming skills (Python).
• Ability to write scientific reports and communicate research results in English.