The artificial intelligence of things (AIOT), which combines the advantages of both artificial intelligence and the technologies of the Internet of Things, has become widespread in recent years. In contrast to typical IoT setups, in which devices collect and transmit data for processing in another location, acquire AIOT devices locally and in real time so that you can make intelligent decisions. This technology has determined extensive applications in intelligent production, security in smart home and in healthcare.
In Smart Home Aiot technology, precise recognition of human activities is of crucial importance. It helps intelligent devices to identify different tasks, e.g. B. cooking and training. Based on this information, the AIOT system can optimize the lighting or change the music automatically, which improves the user experience and at the same time ensures energy efficiency. In this context, WLAN-based motion detection is very promising: WLAN devices are omnipresent, ensure privacy and are usually inexpensive.
Recently, a research team under the direction of Professor Gwanggil Jeon from College of Incheon National University in South Korea has recently developed a new AIOT frame with the name Multiple Specrogram Fusion Network (MSF-Net) for WLAN-based activity detection . Their results were made available online on May 13, 2024 and published in Volume 11, issue 24 of the IEEE Internet of Things Journalon December 15, 2024.
Prof. Jeon explains the motivation behind her research. “As a typical AIOT application, the detection of WLAN-based human activities in smart homes is becoming increasingly popular. However, WLAN-based detection often has an unstable performance due to environmental interferences. Our goal was to overcome this problem.”
In this point of view, the researchers developed the robust deep learning frame MSF-Net, which achieves both rough and fine activity recognition via Channel State Information (CSI). MSF-Net has three main components: a dual stream structure that includes the short-term fourier transformation together with a discrete Wavelet transformation, a transformer and an attention-based fusion branch. While the dual stream structure in CSI determines abnormal information, the transformer is efficiently extracting from the data on a high level. Finally, the fusion branch increases the cross-model fusion.
The researchers carried out experiments to validate the performance of their framework and found that they found remarkable Cohens Kappa scores of 91.82%, 69.76%, 85.91%and 75.66%on Signfi, Widar3. 0, UT-HAR and NTU HAR data sets, data records, data sets, Widar3.0, achieved. respectively. These values ​​underline the superior performance of MSF-Net compared to state-of-the-art techniques for data-based rough and fine activity detection.
“Compared to existing technologies, the multimodal fusion technology has significantly improved the accuracy and efficiency and increases the possibility of practical applications. This research can be used in various areas such as smart homes, rehabilitation medicine and care of older people. For example. Can prevent falls, By analyzing the movements of the user and contributing to improving the quality of life by determining a non-face-to-face health monitoring system, “concludes Prof. Jeon.
Overall, it is expected that activity recognition with WLAN, the convergence technology of IoT and AI in this work is proposed, improve people's lives significantly through everyday convenience and security!