Digital transformation technology innovator Fujitsu said it has successfully developed a new technology to accurately estimate postures of the human body from coarse-grained point cloud data obtained with a conventional millimeter-wave sensor. This technology to help hospital and nursing staff, and patient care takers to visually monitor patients and act in case of emergency such as falling of patients.
While using artificially intelligency based models, accurate estimations of postures of the human body can be made. The movement of patient and the behaviour is analysed using millimeter wave sensors without using privacy invading cameras.
Fujitsu to conduct field trials with hospitals and nursing care facilities to verify the effectiveness and improve the accuracy of the new technology with the aim to offer it as a service to the Japanese market by the end of FY 2023.
Using of millimeter wave sensors in this type of solution is said to be less casting and protect the privacy of patients. However there is an issue with millimeterwave sensors when it comes to accuracy. To address this issue Fujitsu has developed a technology that enables expansion from coarse-grained point cloud data to fine-grained point cloud data necessary for an accurate posture estimation by fusing time series information of point cloud data in a series of movements of the human body. mmWave sensors also help in faster response to emergency situations while ensuring patients’ privacy.
Fujitsu has used common 79 GHz millimeter-wave sensor compliant with the Radio Law of Japan.
Features of the new technology as listed by Fujitsu in its release includes:
1. Point group data extension technology for generating input data suitable for measurements of postures of the human body
Development of a point cloud data extension technology that can be extended to fine-grained point cloud data by selecting point cloud data suitable for the estimation of postures from a large amount of point cloud data that can be acquired by multiple radio waves(3)
2. Large-scale dataset and AI model to measure postures of the human body
Construction of a large-scale data set that combines point cloud data with three dimensional coordinates of human joints based on point cloud data extended to sufficient granularity to estimate the position of the human body to achieve estimations with higher accuracy
Development of an AI model for highly accurate estimations of the position of the human body based on a dataset consisting of behavioral data from about 140 people in about 50 different scenes
3. Actlyzer AI technology for complex and detailed analysis of human behavior
Detailed analysis of various human behaviors (falling after standing up from bed or during walking as well as behavior before and after falling etc.) through utilizing Actlyzer, an AI technology able to analyze complex human behavior by combining approximately 100 types of basic movement data to support nurses and caregivers in visually monitoring patients and achieving a faster response time to emergency situations