How to Improve Height Measurement Accuracy？ #32

Hi,

I have read several of your works related to anthropometric measurement and tried to reproduce some of the ideas in my own setting.

Currently, I am using an RGB-D camera to capture a frontal point cloud of a person in an upright standing pose. The standing pose is relatively natural and relaxed. I do not require the person to stand perfectly straight, but there are also no large body movements.

My goal is to estimate an SMPL model from the frontal RGB-D point cloud and then measure body dimensions such as height. However, in my experiments, the height estimation error is still quite large, and the results are not as accurate as expected.

I would like to ask a few questions:

1. With only a single-view frontal RGB-D point cloud, is there any reliable method to control the height measurement error within 1 cm?
2. Even when depth information is available, why can the SMPL fitting or human body reconstruction still be inaccurate?
3. If the main goal is to improve height measurement accuracy, what attempts should I prioritize? I would really appreciate your advice.

Thank you very much for your time and advice!