Search

The escalating global water usage and the increasing strain on major cities due to water shortages highlights the critical need for efficient water management practices. In water-stressed regions worldwide, significant water wastage is primarily attributed to leakages, inefficient use, and aging infrastructure. Undetected water leakages in buildings' pipelines contribute to the water waste problem. To address this issue, an effective water leak detection method is required. In this paper, we explore the application of edge computing in smart buildings to enhance water management. By integrating sensors and embedded Machine Learning models, known as TinyML, smart water management systems can collect real-time data, analyze it, and make accurate decisions for efficient water utilization. The transition to TinyML enables faster and more cost-effective local decision-making, reducing the dependence on centralized entities. In this work, we propose a solution that can be adapted for effective leakage detection in real-world scenarios with minimum human intervention using TinyML. We follow an approach that is similar to a typical machine learning lifecycle in production, spanning stages including data collection, training, hyperparameter tuning, offline evaluation and model optimization for on-device resource efficiency before deployment. In this work, we considered an existing water leakage acoustic dataset for polyvinyl chloride pipelines. To prepare the acoustic data for analysis, we performed preprocessing to transform it into scalograms. We devised a water leak detection method by applying transfer learning to five distinct Convolutional Neural Network (CNN) variants, which are namely EfficientNet, ResNet, AlexNet, MobileNet V1, and MobileNet V2. The CNN models were found to be able to detect leakages where a maximum testing accuracy, recall, precision, and F1 score of 97.45%, 98.57%, 96.70%, and 97.63%, respectively, were observed using the EfficientNet model. To enable seamless deployment on the Arduino Nano 33 BLE edge device, the EfficientNet model is compressed using quantization resulting in a low inference time of 1932 ms, a peak RAM usage of 255.3 kilobytes, and a flash usage requirement of merely 48.7 kilobytes. [ABSTRACT FROM AUTHOR]

The study of Domain Generalization (DG) has gained considerable momentum in the Machine Learning (ML) field. Human Activity Recognition (HAR) inherently encompasses diverse domains (e.g., users, devices, or datasets), rendering it an ideal testbed for exploring Domain Generalization. Building upon recent work, this paper investigates the application of regularization methods to bridge the generalization gap between traditional models based on handcrafted features and deep neural networks. We apply various regularizers, including sparse training, Mixup, Distributionally Robust Optimization (DRO), and Sharpness-Aware Minimization (SAM), to deep learning models and assess their performance in Out-of-Distribution (OOD) settings across multiple domains using homogenized public datasets. Our results show that Mixup and SAM are the best-performing regularizers. However, they are unable to match the performance of models based on handcrafted features. This suggests that while regularization techniques can improve OOD robustness to some extent, handcrafted features remain superior for domain generalization in HAR tasks. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) has been studied extensively, yet current approaches are not capable of generalizing across different domains (i.e., subjects, devices, or datasets) with acceptable performance. This lack of generalization hinders the applicability of these models in real-world environments. As deep neural networks are becoming increasingly popular in recent work, there is a need for an explicit comparison between handcrafted and deep representations in Out-of-Distribution (OOD) settings. This paper compares both approaches in multiple domains using homogenized public datasets. First, we compare several metrics to validate three different OOD settings. In our main experiments, we then verify that even though deep learning initially outperforms models with handcrafted features, the situation is reversed as the distance from the training distribution increases. These findings support the hypothesis that handcrafted features may generalize better across specific domains. [ABSTRACT FROM AUTHOR]

Physical activity (PA) is globally recognized as a pillar of general health. Step count, as one measure of PA, is a well known predictor of long-term morbidity and mortality. Despite its popularity in consumer devices, a lack of methodological standards and clinical validation remains a major impediment to step count being accepted as a valid clinical endpoint. Previous works have mainly focused on device-specific step-count algorithms and often employ sensor modalities that may not be widely available. This may limit step-count suitability in clinical scenarios. In this paper, we trained neural network models on publicly available data and tested on an independent cohort using two approaches: generalization and personalization. Specifically, we trained neural networks on accelerometer signals from one device and either directly applied them or adapted them individually to accelerometer data obtained from a separate subject cohort wearing multiple distinct devices. The best models exhibited highly accurate step-count estimates for both the generalization (96–99%) and personalization (98–99%) approaches. The results demonstrate that it is possible to develop device-agnostic, accelerometer-only algorithms that provide highly accurate step counts, positioning step count as a reliable mobility endpoint and a strong candidate for clinical validation. [ABSTRACT FROM AUTHOR]

Coronavirus 2019 (COVID-19) has posed a serious threat to the lives and health of the majority of people worldwide. Since the early days of the outbreak, South Korea's government and citizens have made persistent efforts to provide effective prevention against further spread of the disease. In particular, the participation of individual citizens in complying with the necessary code of conduct to prevent spread of the infection, through measures such as social distancing and mask wearing, is as instrumental as the geographical tracking of the trajectory of the infected. In this paper, we propose an activity recognition method based on a wristband equipped with an IR array and inertial measurement unit (IMU) to detect individual compliance with codes of personal hygiene management, such as mask wearing, which are recommended to prevent the spread of infectious diseases. The results of activity recognition were comparatively analyzed by applying conventional machine learning algorithms and convolutional neural networks (CNNs) to the IMU time series and IR array thermal images collected from 25 subjects. When CNN and 24 × 32 thermal images were used, 97.8% accuracy was achieved (best performance), and when 6 × 8 low-resolution thermal images were used, similar performance with 97.1% accuracy was obtained. In the case of using IMU, the performance of activity recognition was lower than that obtained with the IR array, but an accuracy of 93% was achieved even in the case of applying machine learning algorithms, indicating that it is more suitable for wearable devices with low computational capability. [ABSTRACT FROM AUTHOR]

Existing accelerometer-based human activity recognition (HAR) benchmark datasets that were recorded during free living suffer from non-fixed sensor placement, the usage of only one sensor, and unreliable annotations. We make two contributions in this work. First, we present the publicly available Human Activity Recognition Trondheim dataset (HARTH). Twenty-two participants were recorded for 90 to 120 min during their regular working hours using two three-axial accelerometers, attached to the thigh and lower back, and a chest-mounted camera. Experts annotated the data independently using the camera's video signal and achieved high inter-rater agreement (Fleiss' Kappa = 0.96 ). They labeled twelve activities. The second contribution of this paper is the training of seven different baseline machine learning models for HAR on our dataset. We used a support vector machine, k-nearest neighbor, random forest, extreme gradient boost, convolutional neural network, bidirectional long short-term memory, and convolutional neural network with multi-resolution blocks. The support vector machine achieved the best results with an F1-score of 0.81 (standard deviation: ± 0.18 ), recall of 0.85 ± 0.13 , and precision of 0.79 ± 0.22 in a leave-one-subject-out cross-validation. Our highly professional recordings and annotations provide a promising benchmark dataset for researchers to develop innovative machine learning approaches for precise HAR in free living. [ABSTRACT FROM AUTHOR]

Parkinson's disease (PD) is a chronic neurodegenerative condition that affects a patient's everyday life. Authors have proposed that a machine learning and sensor-based approach that continuously monitors patients in naturalistic settings can provide constant evaluation of PD and objectively analyse its progression. In this paper, we make progress toward such PD evaluation by presenting a multimodal deep learning approach for discriminating between people with PD and without PD. Specifically, our proposed architecture, named MCPD-Net, uses two data modalities, acquired from vision and accelerometer sensors in a home environment to train variational autoencoder (VAE) models. These are modality-specific VAEs that predict effective representations of human movements to be fused and given to a classification module. During our end-to-end training, we minimise the difference between the latent spaces corresponding to the two data modalities. This makes our method capable of dealing with missing modalities during inference. We show that our proposed multimodal method outperforms unimodal and other multimodal approaches by an average increase in F 1 -score of 0.25 and 0.09, respectively, on a data set with real patients. We also show that our method still outperforms other approaches by an average increase in F 1 -score of 0.17 when a modality is missing during inference, demonstrating the benefit of training on multiple modalities. [ABSTRACT FROM AUTHOR]

Sensors provide the foundation of many smart applications and cyber–physical systems by measuring and processing information upon which applications can make intelligent decisions or inform their users. Inertial measurement unit (IMU) sensors—and accelerometers and gyroscopes in particular—are readily available on contemporary smartphones and wearable devices. They have been widely adopted in the area of activity recognition, with fall detection and step counting applications being prominent examples in this field. However, these sensors may also incidentally reveal sensitive information in a way that is not easily envisioned upfront by developers. Far worse, the leakage of sensitive information to third parties, such as recommender systems or targeted advertising applications, may cause privacy concerns for unsuspecting end-users. In this paper, we explore the elicitation of age and gender information from gait traces obtained from IMU sensors, and systematically compare different feature engineering and machine learning algorithms, including both traditional and deep learning methods. We describe in detail the prediction methods that our team used in the OU-ISIR Wearable Sensor-based Gait Challenge: Age and Gender (GAG 2019) at the 12th IAPR International Conference on Biometrics. In these two competitions, our team obtained the best solutions amongst all international participants, and this for both the age and gender predictions. Our research shows that it is feasible to predict age and gender with a reasonable accuracy on gait traces of just a few seconds. Furthermore, it illustrates the need to put in place adequate measures in order to mitigate unintended information leakage by abusing sensors as an unanticipated side channel for sensitive information or private traits. [ABSTRACT FROM AUTHOR]

Human falls are a global public health issue resulting in over 37.3 million severe injuries and 646,000 deaths yearly. Falls result in direct financial cost to health systems and indirectly to society productivity. Unsurprisingly, human fall detection and prevention are a major focus of health research. In this article, we consider deep learning for fall detection in an IoT and fog computing environment. We propose a Convolutional Neural Network composed of three convolutional layers, two maxpool, and three fully-connected layers as our deep learning model. We evaluate its performance using three open data sets and against extant research. Our approach for resolving dimensionality and modelling simplicity issues is outlined. Accuracy, precision, sensitivity, specificity, and the Matthews Correlation Coefficient are used to evaluate performance. The best results are achieved when using data augmentation during the training process. The paper concludes with a discussion of challenges and future directions for research in this domain. [ABSTRACT FROM AUTHOR]