📁 Project Overview

🚀 Project Name

New Product 'InBody AI Scale'

🗓 Project Duration

2024.01.29 - 2024.12.19

💼 Object

Consumer Electronics Show 2025, InBody Co., Ltd (Internship)

🧑‍🤝‍🧑 Team Members and Role

Jaeho Kim: PM
Sewon Kim: SW & AI Lead
Seungseo Park
Soohyeon Ga: HW Lead
Yeongjin Kim
Wongoon Lee
Mingwan Seol
Soohong Lim

🎥 Image

💼 Consumer Electronic Show 2025

🎯 Introduction

Inbody AI Scale significantly reduces the measurement time compared to the original InBody models. The original InBody model takes 2 minutes for measurement, but the InBody AI scale completes it in just 40 seconds.

InBody AI scale has two key features.

First, InBody AI Scale require no any input from the user. This was possible through the height measurement and face recognition system using a computer vision AI

Second, InBody AI scale allows users to keep their socks on during measurement.

It was possible through the water spray system. You don’t need to take socks off and don’t need to input the height!

💡 Development

We showed our new product "InBody AI Scale" in Consumer Electornics Show 2025 held in Las Vegas.

As the SW & Vision AI Team Leader, I have developed a computer vision AI solution for the new InBody Scale AI product. This involved researching and enhancing the performance of height estimation algorithms using a low-performance monocular camera and limited computing resources. Additionally, I developed a face verification solution, along with the necessary pipeline and database for it.

Existing height estimation algorithms and papers using vision assume that the user's entire body is within the camera's field of view. However, due to the mechanical characteristics of the InBody device, the user's full body does not fit within the camera's field of view. Our attempt to estimate height under these conditions is unprecedented. In our first experiment, we achieved an average error of 0.6% and a maximum error of 1%. We tackled the ultimate problem of height estimation by breaking it down into sub-problems where computer vision AI can excel: Object Detection and Semantic Segmentation. To achieve this, we designed and utilized a structure that allows the camera to move via motors and rails.