SiMWiSense: Simultaneous Multi-Subject Activity Classification Through Wi-Fi Signals (2304.00057v1)

Abstract: Recent advances in Wi-Fi sensing have ushered in a plethora of pervasive applications in home surveillance, remote healthcare, road safety, and home entertainment, among others. Most of the existing works are limited to the activity classification of a single human subject at a given time. Conversely, a more realistic scenario is to achieve simultaneous, multi-subject activity classification. The first key challenge in that context is that the number of classes grows exponentially with the number of subjects and activities. Moreover, it is known that Wi-Fi sensing systems struggle to adapt to new environments and subjects. To address both issues, we propose SiMWiSense, the first framework for simultaneous multi-subject activity classification based on Wi-Fi that generalizes to multiple environments and subjects. We address the scalability issue by using the Channel State Information (CSI) computed from the device positioned closest to the subject. We experimentally prove this intuition by confirming that the best accuracy is experienced when the CSI computed by the transceiver positioned closest to the subject is used for classification. To address the generalization issue, we develop a brand-new few-shot learning algorithm named Feature Reusable Embedding Learning (FREL). Through an extensive data collection campaign in 3 different environments and 3 subjects performing 20 different activities simultaneously, we demonstrate that SiMWiSense achieves classification accuracy of up to 97%, while FREL improves the accuracy by 85% in comparison to a traditional Convolutional Neural Network (CNN) and up to 20% when compared to the state-of-the-art few-shot embedding learning (FSEL), by using only 15 seconds of additional data for each class. For reproducibility purposes, we share our 1TB dataset and code repository.

• The paper presents SiMWiSense, a novel framework that decentralizes CSI sensing to achieve scalable multi-subject activity classification.
• The paper introduces an innovative FREL algorithm that merges embedding and meta-learning to adapt to new subjects with only 15 seconds of data per class.
• The paper demonstrates robust performance across diverse environments, attaining up to 98% accuracy and outperforming conventional CNN and FSEL benchmarks.

Simultaneous Multi-Subject Activity Classification Using Wi-Fi: Insights into SiMWiSense

In the pursuit of efficient and scalable human activity classification, the presented paper proposes "SiMWiSense"—an innovative framework for simultaneous multi-subject activity recognition leveraging Wi-Fi signals. This work embarks on addressing the core challenges inherent in Wi-Fi based activity classification, primarily the issues of scalability with increased subjects and activities, and the credible generalization across various environments and subjects.

The Framework

The SiMWiSense framework aims for a decentralized detection system, delegating each monitoring device the task of sensing the subject nearest to it. This approach significantly reduces the overall computational complexity by a factor of $P \cdot Q$, where $P$ denotes the number of subjects and $Q$ represents the number of activities. It leverages Channel State Information (CSI) as the foundational sensing data, obtained through several strategically placed Wi-Fi transceivers in differing environments. The central premise is that a CSI monitor can best capture the wireless channel profile of the closest subject, thereby maximizing classification accuracy—a hypothesis validated through proximity tests reported in the paper.

Novel Contributions and Methodology

The authors present a suite of novel methodologies under SiMWiSense, notably a distinct Few-Shot Learning (FSL) algorithm termed "Feature Reusable Embedding Learning (FREL)". FREL combines both embedding learning and meta-learning techniques, showing efficacy in rapidly adapting to new subjects and environments using only 15 seconds of data per class. Through exhaustive experimental evaluations, SiMWiSense showcases up to 98% classification accuracy while outperforming baseline Convolutional Neural Networks (CNNs) by 85% and surpassing state-of-the-art Few-Shot Embedding Learning (FSEL) by 30%, illustrating the significant advancements in accuracy and adaptability.

Experimental Evaluation and Results

The evaluations conducted present robust numerical results, detailing accuracy metrics across various environments—classroom, office, and kitchen—with three subjects performing twenty diversified activities. The achieved classification accuracy baseline on a typical CNN framework set a notable threshold, further improved by FREL, substantiating the authors' claims regarding its proficient generalization capabilities. The empirical demonstration depicts that the SiMWiSense framework effectively sustains performance without substantial degradation, notwithstanding the environmental transition. Additionally, the paper provides a comparative analysis of subcarrier resolution's impact on the system's efficacy, which, although intuitive, provides insightful validation for practical deployment considerations.

Implications and Future Directions

The practical implications of SiMWiSense extend into numerous domains including smart home systems, healthcare monitoring, and security surveillance, where effective multi-subject activity classification can fundamentally enhance system intelligence and responsiveness. The paper hints at several avenues for future exploration, notably expanding subject-matter learning endeavors and fine-tuning the FREL algorithm to further optimize real-time responses and computational efficiency.

Overall, this paper establishes a substantial stride towards more nuanced, scalable, and adaptive systems for real-world Wi-Fi sensing applications. As this line of research unfolds, it promises to bolster the development of intelligent systems with refined granularity in activity recognition, promoting an era of enhanced contextual awareness and autonomy.