Recently, device-free human activity–monitoring systems using commercial Wi-Fi devices have demonstrated a great potential to support smart home environments. These systems exploit Channel State Information (CSI), which represents how human activities–based environmental changes affect the Wi-Fi signals propagating through physical space. However, given that Wi-Fi signals either penetrate through an obstacle or are reflected by the obstacle, there is a high chance that the housing environment would have a great impact on the performance of a CSI-based activity-recognition system. In this context, this paper examines whether and to what extent housing environment affects the performance of the CSI-based activity recognition systems. Activities in daily living (ADL)–recognition systems were implemented in two typical housing environments representative of the United States and South Korea: a wood-frame apartment (Unit A) and a reinforced concrete-frame apartment (Unit B), respectively. The experimental results show that housing environments, combined with various environmental factors (i.e., structural building materials, surrounding Wi-Fi interference, housing layout, and population density), generate a significant difference in the accuracy of the applied CSI-based ADL-recognition systems. This outcome provides insights into how such ADL systems should be configured for various home environments.
With the rapid deployment of indoor Wi-Fi networks, Channel State Information (CSI) has been used for device-free occupant activity recognition. However, various environmental factors interfere with the stable propagation of Wi-Fi signals indoors, which causes temporal variation of CSI data. In this study, we investigated temporal CSI variation in a real-world housing environment and its impact on learning-based occupant activity recognition. The CSI variation over time changes distributions of the CSI data, and the pre-trained model’s accuracy performance becomes degraded during long-term monitoring. In order to address the temporal dependency issue, we developed an effective long-term occupant activity recognition model based on the semi-supervised meta-learning approach. Our model leveraged unlabeled target data with its pseudo labels and synthesized numerous query datasets using mixup-based data augmentation, which generalized the model during training. The model provided an average of 91.09% activity classification accuracy for the target data, which had different statistical characteristics from the source data. This result demonstrates that our model can reliably monitor occupant activities for long-term periods. The dataset presented in this study is available in IEEE DataPort at https://dx.doi.org/10.21227/z10g-vt48.
1 January 2023 in Journal of Computing in Civil Engineering
Manual hazard identification by safety managers in construction has practical challenges because each manager identifies environmental hazards from their perception, which can leave many potential hazards unidentified and consequently lead to accidents at the site. Previous studies have revealed that workers experience loss of body balance (LOB) when exposed to slip, trip, and fall (STF) hazards. This study extended previous studies to identify STF hazards by LOB measurement and collective sensing (i.e., data aggregation) techniques and assumed that STF hazards would cause multiple workers’ LOBs in a given location. First, this study developed an approach to assess each worker’s exposure to STF hazards by LOB analysis. A waist-worn inertial measurement unit sensor was used to extract features of waist movements, which were mapped into a single value to measure LOB scores using the Mahalanobis distance (MD) metric. As an individual worker is exposed to STF hazards, the MD values become larger than without exposure to STF hazards. The developed approach provided an unweighted average recall of 89.13% (without exposures: 90.30%, and with exposures: 87.96%) for detecting individual workers’ exposures to STF hazards in an actual construction site. Then, an approach was developed to visualize the location of STF hazards by allocating multiple workers’ LOB scores into each individual’s Global Positioning System (GPS) data points. The results showed the feasibility of the developed approach to identify STF hazards, potentially helping to prevent STF accidents at construction sites.
1 May 2026 in Journal of Computing in Civil Engineering
Social norm-based interventions on household energy conservation are gaining increasing attention due to their effectiveness and scalability. However, well-established neighbor-based (i.e., geographically based) approaches may be less personally relevant because they overlook individual variations in energy-use behaviors. This study addresses this limitation by providing, to our knowledge, one of the first field evaluations of lifestyle-based normative feedback, tailored to daily energy use patterns identified through clustering of smart meter data. Using a difference-in-differences approach, we compared changes in electricity consumption over one year among 4,562 households assigned to an individual-feedback control group, a neighbor-based feedback group, or a lifestyle-based feedback group, evaluating each treatment relative to the same control group. The lifestyle-based intervention resulted in a 1.963% reduction in annual energy consumption relative to the control (statistically significant, p = 0.029); a 1.437% reduction was observed in the neighbor-based group (marginally significant, p = 0.099). In direct comparison, lifestyle-based feedback achieved a greater reduction than the neighbor-based approach under high-demand conditions, most clearly in winter and among mid-use households (2.722% reduction, p = 0.032). These results demonstrate the importance of personalized normative feedback aligned with user lifestyles and highlight its potential for large-scale implementation, resulting in substantial cumulative energy savings.