Search

Human Activity Recognition (HAR) is a crucial task in numerous applications, including healthcare, smart homes, security, and fitness tracking. This study explores the effectiveness of Long Short-Term Memory (LSTM) networks in accurately recognizing and classifying human activities from sensor data. Leveraging the ability of LSTM to capture temporal dependencies and long-term patterns, we employ a deep learning approach that processes sequential data collected from accelerometers and gyroscopes. Our proposed model demonstrates significant improvements in recognition accuracy. We validate the performance of our approach on benchmark datasets, achieving an accuracy of over 95%. The findings underscore the potential of LSTM networks in advancing HAR systems, offering reliable and precise activity classification that can be integrated into various real-world applications. [ABSTRACT FROM AUTHOR]

Abstract Hand gesture recognition based on sparse multichannel surface electromyography (sEMG) still poses a significant challenge to deployment as a muscle–computer interface. Many researchers have been working to develop an sEMG-based hand gesture recognition system. However, the existing system still faces challenges in achieving satisfactory performance due to ineffective feature enhancement, so the prediction is erratic and unstable. To comprehensively tackle these challenges, we introduce a novel approach: a lightweight sEMG-based hand gesture recognition system using a 4-stream deep learning architecture. Each stream strategically combines Temporal Convolutional Network (TCN)-based time-varying features with Convolutional Neural Network (CNN)-based frame-wise features. In the first stream, we harness the power of the TCN module to extract nuanced time-varying temporal features. The second stream integrates a hybrid Long short-term memory (LSTM)-TCN module. This stream extracts temporal features using LSTM and seamlessly enhances them with TCN to effectively capture intricate long-range temporal relations. The third stream adopts a spatio-temporal strategy, merging the CNN and TCN modules. This integration facilitates concurrent comprehension of both spatial and temporal features, enriching the model’s understanding of the underlying dynamics of the data. The fourth stream uses a skip connection mechanism to alleviate potential problems of data loss, ensuring a robust information flow throughout the network and concatenating the 4 stream features, yielding a comprehensive and effective final feature representation. We employ a channel attention-based feature selection module to select the most effective features, aiming to reduce the computational complexity and feed them into the classification module. The proposed model achieves an average accuracy of 94.31% and 98.96% on the Ninapro DB1 and DB9 datasets, respectively. This high-performance accuracy proves the superiority of the proposed model, and its implications extend to enhancing the quality of life for individuals using prosthetic limbs and advancing control systems in the field of robotic human–machine interfaces.

Human Activity Recognition (HAR) plays a crucial role in computer vision and signal processing, with extensive applications in domains such as security, surveillance, and healthcare. Traditional machine learning (ML) approaches for HAR have achieved commendable success but face limitations such as reliance on handcrafted features, sensitivity to noise, and challenges in handling complex temporal dependencies. These limitations have spurred interest in deep learning (DL) and hybrid models that address these issues more effectively. Thus, DL has emerged as a powerful approach for HAR, surpassing the performance of traditional methods. In this paper, a multi-modal hybrid hierarchical classification approach is proposed. It combines DL transformers with traditional ML techniques to improve both the accuracy and efficiency of HAR. The proposed classifier is evaluated based on the different activities of four widely used benchmark datasets: PAMAP2, CASAS, UCI HAR, and UCI HAPT. The experimental results demonstrate that the hybrid hierarchical classifier achieves remarkable accuracy rates of 99.69%, 97.4%, 98.7%, and 98.6%, respectively, outperforming traditional classification methods and significantly reducing training time compared to sequential LSTM models. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) is crucial in applications such as smart homes, interactive games, surveillance, security, and healthcare. In recent years, Channel State Information (CSI) data extracted from Wi-Fi signals has garnered significant interest for applications in HAR. This interest stems from CSI's several advantages, including its immunity to illumination variations and environmental disturbances, and the elimination of the need for wearable devices. Despite being widely used, existing HAR system's performance suffers when used in new environments without system improvement or retraining. This constraint can be overcome by gathering and annotating data from various locations, and then re-training the system. However, it is far from ideal from the privacy perspective, as the training algorithms access the data from different privacy-sensitive environments. This motivates us to design a reliable and robust privacy-preserving HAR system. In this work, we introduce a Differentially Private Principal Component-based Wavelet Convolutional Neural Network (DP-PCWCNN) that offers accurate and robust HAR performance across different environments, while preserving strict privacy constraints. We evaluate the performance of our proposed algorithm on two publicly available real datasets and demonstrate that our proposed system closely approximates the non-private system's performance for some parameter choices. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) is a crucial task in various applications, including healthcare, fitness, and the military. Deep learning models have revolutionized HAR, however, their computational complexity, particularly those involving BiLSTMs, poses significant challenges for deployment on resource-constrained devices like smartphones. While BiLSTMs effectively capture long-term dependencies by processing inputs bidirectionally, their high parameter count and computational demands hinder practical applications in real-time HAR. This study investigates the approximation of the computationally intensive BiLSTM component in a HAR model by using a combination of alternative model components and data flipping augmentation. The proposed modifications to an existing hybrid model architecture replace the BiLSTM with standard and residual LSTM, along with convolutional networks, supplemented by data flipping augmentation to replicate the context awareness typically provided by BiLSTM networks. The results demonstrate that the residual LSTM (ResLSTM) model achieves superior performance while maintaining a lower computational complexity compared to the traditional BiLSTM model. Specifically, on the UCI-HAR dataset, the ResLSTM model attains an accuracy of 96.34% with 576,702 parameters, outperforming the BiLSTM model's accuracy of 95.22% with 849,534 parameters. On the WISDM dataset, the ResLSTM achieves an accuracy of 97.20% with 192,238 parameters, compared to the BiLSTM's 97.23% accuracy with 283,182 parameters, demonstrating a more efficient architecture with minimal performance trade-off. For the KU-HAR dataset, the ResLSTM model achieves an accuracy of 97.05% with 386,038 parameters, showing comparable performance to the BiLSTM model's 98.63% accuracy with 569,462 parameters, but with significantly fewer parameters. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) has emerged as a vital measure of quality of life, holding significant implications for human health. The need for effective mobility monitoring in diverse settings, both indoors and outdoors, necessitates the development of scientific and technological tools. To broaden accessibility, wearable devices like smartphones and smartwatches are commonly employed, leveraging sensor data analysis for valuable insights into user activities. HAR, treated as a classification task, involves training a classifier using sensor data and corresponding activity labels. The classifier aims to automatically recognize and classify various activities in future instances. However, the performance of a generic HAR model trained on a diverse population diminishes significantly when applied to a specific user due to inter-subject variability, encompassing variations in activity patterns, behavioral status, gait, and posture. To address this challenge, we propose an innovative auto-supervised domain adaptation approach based on particle filter theory, aiming to automatically construct a personalized HAR classifier. Our approach integrates multiple steps inspired by the particle filter formalism, enabling iterative approximation of the target distribution through temporal samples to personalize the HAR model for the specific user. In extensive experiments on public HAR datasets, we emphasize the critical role of personalization when deploying an HAR classifier for a new user. The results demonstrate that our framework significantly enhances the accuracy of HAR for new users compared to a non-personalized model, achieving an average improvement of 50% across most datasets. Furthermore, we implement our personalized HAR model on an embedded wearable device, enhancing its accessibility for real-world applications. [ABSTRACT FROM AUTHOR]

As Canada's population of older adults rises, the need for aging-in-place solutions is growing due to the declining quality of long-term-care homes and long wait times. While the current standards include questionnaire-based assessments for monitoring activities of daily living (ADLs), there is an urgent need for advanced indoor localization technologies that ensure privacy. This study explores the use of Ultra-Wideband (UWB) technology for activity recognition in a mock condo in the Glenrose Rehabilitation Hospital. UWB systems with built-in Inertial Measurement Unit (IMU) sensors were tested, using anchors set up across the condo and a tag worn by patients. We tested various UWB setups, changed the number of anchors, and varied the tag placement (on the wrist or chest). Wrist-worn tags consistently outperformed chest-worn tags, and the nine-anchor configuration yielded the highest accuracy. Machine learning models were developed to classify activities based on UWB and IMU data. Models that included positional data significantly outperformed those that did not. The Random Forest model with a 4 s data window achieved an accuracy of 94%, compared to 79.2% when positional data were excluded. These findings demonstrate that incorporating positional data with IMU sensors is a promising method for effective remote patient monitoring. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) is pivotal in advancing applications ranging from healthcare monitoring to interactive gaming. Traditional HAR systems, primarily relying on single data sources, face limitations in capturing the full spectrum of human activities. This study introduces a comprehensive approach to HAR by integrating two critical modalities: RGB imaging and advanced pose estimation features. Our methodology leverages the strengths of each modality to overcome the drawbacks of unimodal systems, providing a richer and more accurate representation of activities. We propose a two-stream network that processes skeletal and RGB data in parallel, enhanced by pose estimation techniques for refined feature extraction. The integration of these modalities is facilitated through advanced fusion algorithms, significantly improving recognition accuracy. Extensive experiments conducted on the UTD multimodal human action dataset (UTD MHAD) demonstrate that the proposed approach exceeds the performance of existing state-of-the-art algorithms, yielding improved outcomes. This study not only sets a new benchmark for HAR systems but also highlights the importance of feature engineering in capturing the complexity of human movements and the integration of optimal features. Our findings pave the way for more sophisticated, reliable, and applicable HAR systems in real-world scenarios. [ABSTRACT FROM AUTHOR]

Recently, wireless sensing techniques have been widely used for Internet of Things (IoT) applications. Unlike traditional device-based sensing, wireless sensing is contactless, pervasive, low cost, and non-invasive, making it highly suitable for relevant IoT applications. However, most existing methods are highly dependent on high-quality datasets, and the minority class will not achieve a satisfactory performance when suffering from a class imbalance problem. In this article, we propose a time–frequency semantic generative adversarial network framework (i.e., TFSemantic) to address the imbalanced classification problem in human activity recognition using radio frequency (RF) signals. Specifically, the TFSemantic framework can learn semantic features from the minority classes and then generate high-quality signals to restore data balance. It includes a data pre-processing module, a semantic extraction module, a semantic distribution module, and a data augmenter module. In the data pre-processing module, we process four different RF datasets (i.e., WiFi, RFID, UWB, and mmWave). We also develop Fourier semantic feature convolution and attention semantic feature embedding methods for the semantic extraction module. A discrete wavelet transform is utilized for reconstructed RF samples in the semantic distribution module. In data augmenter module, we design an associated loss function to achieve effective adversarial training. Finally, we validate the effectiveness of the proposed TFSemantic framework using different RF datasets, which outperforms several state-of-the-art methods. [ABSTRACT FROM AUTHOR]

This study explored an indoor system for tracking multiple humans and detecting falls, employing three Millimeter-Wave radars from Texas Instruments. Compared to wearables and camera methods, Millimeter-Wave radar is not plagued by mobility inconveniences, lighting conditions, or privacy issues. We conducted an initial evaluation of radar characteristics, covering aspects such as interference between radars and coverage area. Then, we established a real-time framework to integrate signals received from these radars, allowing us to track the position and body status of human targets non-intrusively. Additionally, we introduced innovative strategies, including dynamic Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering based on signal SNR levels, a probability matrix for enhanced target tracking, target status prediction for fall detection, and a feedback loop for noise reduction. We conducted an extensive evaluation using over 300 min of data, which equated to approximately 360,000 frames. Our prototype system exhibited a remarkable performance, achieving a precision of 98.9% for tracking a single target and 96.5% and 94.0% for tracking two and three targets in human-tracking scenarios, respectively. Moreover, in the field of human fall detection, the system demonstrates a high accuracy rate of 96.3%, underscoring its effectiveness in distinguishing falls from other statuses. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Donghua University (English Edition) is the property of Journal of Donghua University Editorial Board and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Human activity recognition (HAR) is one of the most interesting application for machine learning models running on low-cost and low-power devices, such as microcontrollers (MCUs). As a matter of fact, MCUs are often dedicated to performing inference on their own acquired data, and any form of model training and update is delegated to external resources. We consider this mainstream paradigm a severe limitation, especially when privacy concerns prevent data sharing, thus model personalization, which is universally recognized as beneficial in HAR. In this letter, we present our HAR solution where MCUs can directly fine-tune a deep learning model using locally acquired data. In particular, we enable training functionalities for 1-D convolutional neural networks (CNNs) on STM32 microcontrollers and provide a software tool to estimate the memory and computational resources required to accomplish model personalization. [ABSTRACT FROM AUTHOR]

Abstract The detection of Activities of Daily Living (ADL) holds significant importance in a range of applications, including elderly care and health monitoring. Our research focuses on the relevance of ADL detection in elderly care, highlighting the importance of accurate and unobtrusive monitoring. In this paper, we present a novel approach that that leverages smartphone data as the primary source for detecting ADLs. Additionally, we investigate the possibilities offered by ambient sensors installed in smart home environments to complement the smartphone data and optimize the ADL detection. Our approach uses a Long Short-Term Memory (LSTM) model. One of the key contributions of our work is defining ADL detection as a multilabeling problem, allowing us to detect different activities that occur simultaneously. This is particularly valuable since in real-world scenarios, individuals can perform multiple activities concurrently, such as cooking while watching TV. We also made use of unlabeled data to further enhance the accuracy of our model. Performance is evaluated on a real-world collected dataset, strengthening reliability of our findings. We also made the dataset openly available for further research and analysis. Results show that utilizing smartphone data alone already yields satisfactory results, above 50% true positive rate and balanced accuracy for all activities, providing a convenient and non-intrusive method for ADL detection. However, by incorporating ambient sensors, as an additional data source, one can improve the balanced accuracy of the ADL detection by 7% and 8% of balanced accuracy and true positive rate respectively, on average.

Human activity recognition (HAR) is an important research area in both machine learning and human–computer interactions. Unfortunately, it remains an extremely difficult process owing to unsolvable issues, such as sensor movement, sensor positioning, crowded background, and inherent diversity in task performances by distinct humans. In this study, we developed an ensemble of classification models for the HAR. The proposed HAR has four working phases—preprocessing, segmentation, feature extraction, and classification. The pre-processing phase includes processes such as frame conversion and contrast enhancement. We developed an improved balanced iterative reducing and clustering utilising hierarchies (BIRCH) algorithm, that provides efficient segmentation by utilizing only minimal resources. These segmented images are subjected to feature extraction, in which grey level co-occurrence matrix (GLCM) features, and improved local gradient threshold pattern (LGTP) features are extracted along with conventional bag of visual words (BoVW) to provide better results. An ensemble classification model with classifiers such as Bi-GRU, CNN, and LSTM was developed in this study to provide an accurate classification. To enhance the performance of the proposed model, we developed a blue monkey standardized aquila optimization (BMSAO) approach. Conventional techniques are contrasted with the proposed framework. The proposed mechanism was found to have higher efficiency in HAR after it was experimentally evaluated. [ABSTRACT FROM AUTHOR]

Walking is a complex task that requires consistent practice to master, and it involves the synchronisation between the lower limbs and the brain, making it challenging. While bipedal robots have been developed to mimic human walking, they must achieve an efficient gait due to structural differences and walking challenges. This study aims to produce a more human-like walk by analysing human lower extremity activities. To capture the bipedal robot locomotion learning process, an ensemble classifier based on deep learning is introduced to recognise human lower activities. A publicly available UC Irvine Machine Learning Repository (UCI) dataset on surface electromyography (sEMG) signal for the lower extremity of 11 fit participants and 11 participants with knee disorders for sitting while performing knee extension, walking, and standing while performing knee flexion is used. A hybrid ensemble of deep learning models comprising long short-term memory and convolution neural network is employed to classify activities, with reported average accuracies of 98.8%, 98.3%, and 99.3% for healthy subjects for sitting, standing and walking, respectively. Moreover, the ensemble model reported average accuracies of 98.2%, 98.1%, and 99.0% for individuals with knee pathology. Notably, this study holds promising significance, as it has yielded a considerable enhancement in performance as opposed to state-of-the-art work. The applications of this work are diverse and include improving postural stability in elderly subjects, aiding in the rehabilitation of patients recovering from stroke and trauma, generating walking trajectories for robots in complex environments, and reconstructing walking patterns in individuals with impairments. [ABSTRACT FROM AUTHOR]

The increase in the use of electronic devices and the high rate of data stream production such as video reveals the importance of analyzing the content of such data. Content analysis of video data for human activity recognizing (HAR) has a significant application in the science of machine vision. So far, vast studies have been conducted to HAR subject. Also, despite many challenges in the research field of video data content analysis, previous researchers have proposed many effective methods in field of human activity recognition. However, the literature reveals lacking of proper context for identification, analysis and evaluation of the HAR methods and challenges in a coherent and uniform form to achieve a macro vision of the HAR subject. Hence, it seems necessary to present a comprehensive and comparative analytical review regarding the HAR on video data relying on methods and challenges. The novelty of this research is to present a comparative analytical framework called HAR-CO, which provide a macro vision, coherent structure and deeper understanding concerning to the HAR. The HAR-CO consists of three main parts. Firstly, categorizing the HAR methods in a coherent and structured way based on data collection hardware. Secondly, categorizing HAR challenges in a systematic based on the sensor attachment. Thirdly, a comparative analytical evaluation of each class of HAR approaches according to challenges toward researchers. We think that the HAR-CO framework can serve as road map and guide to select a more appropriate of HAR methods and provide new research directions by researchers. [ABSTRACT FROM AUTHOR]

Recent advancements in the Internet of Things (IoT) wearable devices such as wearable inertial sensors have increased the demand for precise human activity recognition (HAR) with minimal computational resources. The wavelet transform, which offers excellent time-frequency localization characteristics, is well suited for HAR recognition systems. Selecting a mother wavelet function in wavelet analysis is critical, as optimal selection improves the recognition performance. The activity time signals data have different periodic patterns that can discriminate activities from each other. Therefore, selecting a mother wavelet function that closely resembles the shape of the recognized activity's sensor (inertial) signals significantly impacts recognition performance. This study uses an optimal mother wavelet selection method that combines wavelet packet transform with the energy-to-Shannon-entropy ratio and two classification algorithms: decision tree (DT) and support vector machines (SVM). We examined six different mother wavelet families with different numbers of vanishing points. Our experiments were performed on eight publicly available ADL datasets: MHEALTH, WISDM Activity Prediction, HARTH, HARsense, DaLiAc, PAMAP2, REALDISP, and HAR70+. The analysis demonstrated in this paper can be used as a guideline for optimal mother wavelet selection for human activity recognition. [ABSTRACT FROM AUTHOR]

The detection of Activities of Daily Living (ADL) holds significant importance in a range of applications, including elderly care and health monitoring. Our research focuses on the relevance of ADL detection in elderly care, highlighting the importance of accurate and unobtrusive monitoring. In this paper, we present a novel approach that that leverages smartphone data as the primary source for detecting ADLs. Additionally, we investigate the possibilities offered by ambient sensors installed in smart home environments to complement the smartphone data and optimize the ADL detection. Our approach uses a Long Short-Term Memory (LSTM) model. One of the key contributions of our work is defining ADL detection as a multilabeling problem, allowing us to detect different activities that occur simultaneously. This is particularly valuable since in real-world scenarios, individuals can perform multiple activities concurrently, such as cooking while watching TV. We also made use of unlabeled data to further enhance the accuracy of our model. Performance is evaluated on a real-world collected dataset, strengthening reliability of our findings. We also made the dataset openly available for further research and analysis. Results show that utilizing smartphone data alone already yields satisfactory results, above 50% true positive rate and balanced accuracy for all activities, providing a convenient and non-intrusive method for ADL detection. However, by incorporating ambient sensors, as an additional data source, one can improve the balanced accuracy of the ADL detection by 7% and 8% of balanced accuracy and true positive rate respectively, on average. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) is focused on Activities of Daily Living and developed in the health and human security fields. The HAR concept was introduced in previous research using multi-sensor devices. In their implementation, wearable devices require computational and real-time environmental limitations. This paper proposed a new approach for HAR using a machine learning-based single-sensor accelerometer. This research aimed to determine the performance of machine learning in HAR using three Feature Subsets: Feature Subset Signal Vector Magnitude (SMA), Feature Subset Fast Fourier Transform (FFT), and Feature Subset Value-Crossing. In features selection, ANOVA was used to reduce feature dimensionality. The experimental results have been assessed using the confusion matrix to prove that the proposed model can achieve an optimal accuracy of 0.97, higher than several state-of-the-art approaches. The optimal sensitivity and specificity values have been 0.98 and 0.99 and are partially higher than previous studies using similar testing scenarios. [ABSTRACT FROM AUTHOR]

Innovative monitoring solutions utilizing radar have been developed to meet various safety needs. Radar technologies, equipped with unique features, have proven to be effective tools for human activity recognition in diverse environments. Radar complements other sensors by ensuring privacy while recording movement, particularly in situations where video cameras may be intrusive. Deep learning techniques, including convolutional neural networks (CNN) and recurrent neural networks (RNN), enhance the recognition of human behaviors based on radar, providing solutions for fall detection and motion tracking. Transfer learning addresses issues related to data scarcity, while data fusion models integrate information from multiple sources. Advances in machine learning, data processing, and GPU speed contribute to the increased efficiency of radar technology for indoor monitoring, opening possibilities for more advanced applications. The aim of this article is to discuss the evolving field of human behavior analysis using radar and LiDAR. To this end, the literature in this field has been reviewed. In this study, we focus not only on the analysis of human behavior using radar but also on video observations. We review methods of human activity recognition using LiDAR. We discuss potential directions for development, draw conclusions, and outline future research in the area of Human Activity Recognition (HAR).

Sensor-based datasets are extensively utilized in human-computer interaction (HCI) and medical applications due to their portability and strong privacy features. Many researchers have developed sensor-based human activity recognition (HAR) systems to increase recognition performance. However, existing systems still face challenges in achieving satisfactory performance due to insufficient time-varying features and gradient explosion issues. To address these challenges, we proposed a multi-stream temporal convolutional network (TCN)-based approach for time-varying feature extraction and feature selection to recognize human activity (HA) from sensor datasets. The proposed model effectively extracts and emphasizes the spatial-temporal features of various human activities based on a 4-stream model. Each stream uses TCN to extract time-varying features and enhances them using an appropriate integration module. The first stream extracts fine-grained temporal features with TCN. The second and third streams integrate TCN features with LSTM, applying pre-integration and post-integration, respectively. The fourth stream uses CNN for spatial features and TCN for enhancing temporal features. The concatenation of the 4-stream features captures complex dependencies, improving the model’s understanding of prolonged activities. In addition, we proposed a modified effective channel attention network (ECA-Net) that assigns higher dimensionality weight to lower dimensionality, enabling the proposed model to learn and recognise human activities effectively despite their complex patterns. Evaluations on the WISDM, PAMAP2, USC-HAD, Opportunity UCI, and UCI-HAR datasets showed accuracy improvements of 1.12%, 1.99%, 1.30%, 5.72%, and 0.38%, respectively, over state-of-the-art systems. The high-performance accuracy of the proposed model demonstrates its superiority, with implications for improving prosthetic limb functionality and advancing robotics human-machine interfaces. Our data preprocessing approach, deep learning model code, and dataset information are available at the following link: https://github.com/musaru/HAR_Sensor.

Human Activity Recognition (HAR) is becoming increasingly important in the fast-evolving landscapes of wearable sensors, smart applications, and the Internet of Things (IoT) paradigms. HAR is rapidly gaining importance, especially in health monitoring, elderly and infant care, fitness tracking, and security. Machine learning (ML) and Deep Learning (DL) methods are used significantly for HAR problems. ML and DL methods often face four significant problems. Firstly, they lack the adaptability to evolve network architectures dynamically, which is vital in complex tasks like HAR. Secondly, classical ML and DL methods could fall short in comprehensively navigating potential solution spaces. Thirdly, finding optimal hyperparameters is a computationally expensive process, and lastly, expert knowledge is required to configure the hyperparameters. This paper proposes the NeuroHAR, a transformative neuroevolutionary method for HAR, to address these problems. NeuroHAR integrates feedforward deep neural networks (FDNN) with evolutionary algorithms. The highlights of NeuroHAR include its dynamic optimization of network architectures and hyperparameters. It is simple to configure and computationally efficient. Due to its adaptable design, it offers a more flexible and robust solution that handles task complexities better than traditional methods. Results are promising, which is evidence of the effectiveness of the proposed NeuroHAR, which outperformed the explicit contender, the state-of-the-art Grid Search approach. NeuroHAR and Grid Search evaluated 900 and 1080 models in Case I. Even with broader hyperparameter ranges in Case II, NeuroHAR still executed 900 models, whereas Grid Search would need over 2 billion models, which proves the computational efficiency of NeuroHAR. Additionally, for HARTH and HAR70Plus imbalanced datasets, the NeuroHAR model achieved a higher prediction accuracy of 89.91% versus Grid Search’s 84.04%. NeuroHAR can facilitate advanced monitoring and analytics, which are crucial for health monitoring, elderly care, and urban management powered by wearable sensors, IoT, and smart applications.

Automated Human Activity Recognition (HAR) stems from the requirement to seamlessly integrate technology into daily life, to enhance user experience, improve healthcare, provide improved operations, ensure safety, deliver data-driven insights, and address various real-world challenges. However, unscripted Human activity faces challenges that must be understood, and require advances in sensor technology and machine learning models. This paper presents an Active Transfer Learning (ATL) approach for real-time HAR using mobile sensor data. Unlike traditional methods, our approach accounts for both the physical and habitual constraints of individuals. Existing works make an unrealistic assumption of an omniscient oracle while using the same datasets for both training and testing of the models, which makes them impractical for industry requirements. Our proposed approach addresses challenges in existing HAR algorithms, proposing a methodology to adapt models to the real-world datasets while training and testing on cross datasets. We have tailored an existing Entropy and Memory Maximization algorithm to work in a real-time environment while considering user constraints. Primarily trained in a well-labeled controlled environment dataset, we introduce noise injection to prevent the model from overfitting and enhance its generalization for scarcely labeled real-world datasets. Evaluations on publicly available datasets demonstrate our approach achieves 80% - 90% accuracy, outperforming the base algorithm accuracy of 12% - 14%. Importantly, our proposed technique outperforms with limited labeled data, making it valuable for real-time scenarios where labeling is sparse. This research advances HAR in real-world settings, offering improved accuracy and adaptability.

Human Activity Recognition (HAR) based on Inertial Measurement Unit (IMU) has become increasingly important in health and fitness applications. These systems can continuously and cost-effectively monitor human activity, regardless of the surrounding environment. However, the current dominant trend in HAR uses black box flat classification (FC) methods, which lack interpretability and do not consider the natural hierarchical relationship between activity classes. Such systems often achieve greater accuracy at the cost of increased complexity and are not suitable for critical decision-making applications. This paper proposes an Adaptive Hierarchical Decision Tree (AHDT) HAR system that recognizes human activities based on IMU measurements along with quasi-stationary inclination feature extraction. The proposed system generates a global classifier that classifies human activities according to a tree taxonomy structure. This approach maintains interpretability while considering the fundamental signal data features embedded in the natural hierarchical representation of the activity classes. In addition to the commonly used flat classification accuracy measures, we applied modified hierarchical accuracy measures to assess the exclusive characteristics of hierarchical relationships between the classes. We used seven publicly available datasets to evaluate the proposed system and compared its performance with other tree-based classifiers, including Random Forest, Gradient Boosting, XGBoost, and AdaBoost classifiers. Our results demonstrated that the AHDT system significantly improves the recognition performance of fine-grained activities and offers a balance between lower complexity and higher interpretability. Overall, the proposed AHDT system provides an interpretable and practical approach to HAR that can be valuable in critical decision-making applications.

This research explores the potential of technologies in human activity recognition among the elderly population. More precisely, using sensor data and implementing Active Learning (AL), Machine Learning (ML), and Deep learning (DL) techniques for elderly activity recognition. Moreover, the study leverages the HAR70+ dataset, providing insight into the daily activities of older individuals and AL-based ML and DL techniques to construct predictive models for these activities. The findings have implications for proactive and personalized elderly care, representing an approach to improving prediction performance in this domain. The research experiments are presented systematically, summarizing the outcomes of various machine-learning models across three iterative experiments. This research explored a diverse array of ML algorithms, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Logistic Regression (LR), K-Nearest Neighbors (KNN), Stochastic Gradient Descent (SGB) and DL methods such as Deep Neural Networks (DNN) and Long Short-Term Memory networks (LSTM) for experimentation. This research trained models on 7 activities: walking, shuffling, climbing stairs (up and down), standing, sitting, and lying down, and 4 activities separately: standing, sitting, walking, and lying down, using the same classifiers. Results reveal that LSTM achieved the best accuracy of 0.95% for 7 activities and 0.96% using RF on 4 actives, showing the potential of DL and ML techniques, particularly when integrated with AL, to enhance activity recognition rate, patient care, optimize medication strategies and improve the well-being of elderly individuals. Hence, the findings presented in this study have showcased the potential to enhance the quality of life for seniors using the blend of ML, DL and AL.

In recent years, human activity recognition (HAR) has gained importance in several domains such as surveillance, recognizing indoor and outdoor activities, and providing active and assisted living environments in smart homes and healthcare services. In all these scenarios, audio-, video- and image-based processing algorithms have been applied as well as systems using wearable sensors. This scoping review focuses on audio- and video-based activity recognition systems for healthcare applications. We provide a comprehensive overview of these systems and technologies and discuss their complexity, performance, robustness, stage of development, scalability as well as achievable privacy and security levels. Additionally, we present and discuss datasets that are designed for the evaluation of these activity recognition systems. Although a number of robust approaches have already been proposed, they still pose challenges when it comes to integrating them in larger systems or into clinical practice. We identify challenges for application of audio- and video-based HAR systems in real-word healthcare scenarios, draw recommendations and conclusions based on comparisons of existing approaches, and analyze future trends.

Human activity recognition (HAR) based on wearable sensors has emerged as a low-cost key-enabling technology for applications such as human–computer interaction and healthcare. In wearable sensor-based HAR, deep learning is desired for extracting human active features. Due to the spatiotemporal dynamic of human activity, a special deep learning network for recognizing the temporal continuous activities of humans is required to improve the recognition accuracy for supporting advanced HAR applications. To this end, a residual multifeature fusion shrinkage network (RMFSN) is proposed. The RMFSN is an improved residual network which consists of a multi-branch framework, a channel attention shrinkage block (CASB), and a classifier network. The special multi-branch framework utilizes a 1D-CNN, a lightweight temporal attention mechanism, and a multi-scale feature extraction method to capture diverse activity features via multiple branches. The CASB is proposed to automatically select key features from the diverse features for each activity, and the classifier network outputs the final recognition results. Experimental results have shown that the accuracy of the proposed RMFSN for the public datasets UCI-HAR, WISDM, and OPPORTUNITY are 98.13%, 98.35%, and 93.89%, respectively. In comparison with existing advanced methods, the proposed RMFSN could achieve higher accuracy while requiring fewer model parameters. [ABSTRACT FROM AUTHOR]

Human Activity Recognition (HAR) has several applications in healthcare, security, and assisted living systems used in smart homes. The main aim of these applications or systems is to classify body movement read from the built in sensors such as accelerometers and gyroscopes. Some actions could be performed in response to the output of these HAR systems. The number of smartphone users increases, whereas the sensors are widely available in different sizes and shapes (internal or external sensors). Recent advances in sensor technology and machine learning have led researchers to conduct studies on sensor technology such as HAR. HAR systems typically use a combination of sensors, such as accelerometers, gyroscopes, and cameras, to collect images or signal data that can be classified by machine learning algorithms. HAR research has focused on several key challenges including dealing with variability in sensor data, handling missing data or noise, and dealing with large amounts of sensor-generated data. In this work, several machine learning algorithms were tested in predefined settings using the KU-HAR dataset in a series of experiments. Subsequently, various performance metrics were calculated to assess the chosen algorithms' performance. The experimental findings showed that the LightGBM classifier surpassed the other machine learning algorithms in performance metrics, such as accuracy, F1 score, precision, and recall. Although Gradient Boosting has lengthy training time, the other classifiers complete their training in an acceptable time period. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) has a wide range of applications in medical, military, smart home, and other fields. Feature extraction using traditional machine learning methods is challenging and imprecise. Aiming to address the aforementioned issues and leveraging the timing characteristics of the sensor, we propose a DRSN-GRU model that integrates the CBAM attention mechanism. The problem of gradient explosion and vanishing, which was caused by increasing network depth in traditional sequential models was effectively avoided. The parallel structure allows both branches to be given the same priority. The deep residual shrinkage network (DRSN) was used to extract the deep spatial characteristics of the data, while the gated recurrent unit (GRU) was used to learn the characteristics of the activity samples over time series. The two channels simultaneously extract features of different dimensions from the samples and distribute the weight of these features using the CBAM module. Finally, the end-to-end human activity recognition is achieved through the Softmax layer for classification. The WISDM public data set was used for verification, and the model achieved an average accuracy of 97. 6%. This demonstrates a superior recognition effect compared to the traditional machine learning and neural network models proposed by previous researchers. [ABSTRACT FROM AUTHOR]

Human activity recognition (HAR) is a challenging issue in several fields, such as medical diagnosis. Recent advances in the accuracy of deep learning have contributed to solving the HAR issues. Thus, it is necessary to implement deep learning algorithms that have high performance and greater accuracy. In this paper, a gated recurrent unit (GRU) algorithm is proposed to classify human activities. This algorithm is applied to the Wireless Sensor Data Mining (WISDM) dataset gathered from many individuals with six classes of various activities – walking, sitting, downstairs, jogging, standing, and upstairs. The proposed algorithm is tested and trained via a hyper-parameter tuning method with TensorFlow framework to achieve high accuracy. Experiments are conducted to evaluate the performance of the GRU algorithm using receiver operating characteristic (ROC) curves and confusion matrices. The results demonstrate that the GRU algorithm provides high performance in the recognition of human activities. The GRU algorithm achieves a testing accuracy of 97.08%. The rate of testing loss for the GRU is 0.221, while the precision, sensitivity, and F1-score for the GRU are 97.11%, 97.09%, and 97.10%, respectively. Experimentally, the area under the ROC curves (AUCS) is 100%. [ABSTRACT FROM AUTHOR]

In our research, we propose a model that leverages transfer learning and active learning techniques to accumulate knowledge and effectively solve complex problems in the field of artificial intelligence. This model operates within a parallel learning paradigm, aiming to mimic the continuous learning and improvement observed in human beings. To facilitate knowledge accumulation, we introduce a convolutional deep classification auto encoder that extracts spatially localized features from images. This enhances the model's ability to extract relevant information. Additionally, we propose a learning classification system based on a code fragment, enabling effective representation and transfer of knowledge across different domains. Our research culminates in a theoretical and practical prototype for active learning-based knowledge extraction in various living organisms, including humans, plants, and animals. This knowledge extraction is achieved through image-based learning transfer, focusing on advancing activity recognition in image processing. Experimental results confirm that our method outperforms both baseline approaches and state-of-the-art convolutional neural network methods, underscoring its effectiveness and potential. [ABSTRACT FROM AUTHOR]

Human capability to perform routine tasks declines with age and age-related problems. Remote human activity recognition (HAR) is beneficial for regular monitoring of the elderly population. This paper addresses the problem of the continuous detection of daily human activities using a mm-wave Doppler radar. In this study, two strategies have been employed: the first method uses un-equalized series of activities, whereas the second method utilizes a gradient-based strategy for equalization of the series of activities. The dynamic time warping (DTW) algorithm and Long Short-term Memory (LSTM) techniques have been implemented for the classification of un-equalized and equalized series of activities, respectively. The input for DTW was provided using three strategies. The first approach uses the pixel-level data of frames (UnSup-PLevel). In the other two strategies, a convolutional variational autoencoder (CVAE) is used to extract Un-Supervised Encoded features (UnSup-EnLevel) and Supervised Encoded features (Sup-EnLevel) from the series of Doppler frames. The second approach for equalized data series involves the application of four distinct feature extraction methods: i.e., convolutional neural networks (CNN), supervised and unsupervised CVAE, and principal component Analysis (PCA). The extracted features were considered as an input to the LSTM. This paper presents a comparative analysis of a novel supervised feature extraction pipeline, employing Sup-ENLevel-DTW and Sup-EnLevel-LSTM, against several state-of-the-art unsupervised methods, including UnSUp-EnLevel-DTW, UnSup-EnLevel-LSTM, CNN-LSTM, and PCA-LSTM. The results demonstrate the superiority of the Sup-EnLevel-LSTM strategy. However, the UnSup-PLevel strategy worked surprisingly well without using annotations and frame equalization. [ABSTRACT FROM AUTHOR]

Human capability to perform routine tasks declines with age and age-related problems. Remote human activity recognition (HAR) is beneficial for regular monitoring of the elderly population. This paper addresses the problem of the continuous detection of daily human activities using a mm-wave Doppler radar. In this study, two strategies have been employed: the first method uses un-equalized series of activities, whereas the second method utilizes a gradient-based strategy for equalization of the series of activities. The dynamic time warping (DTW) algorithm and Long Short-term Memory (LSTM) techniques have been implemented for the classification of un-equalized and equalized series of activities, respectively. The input for DTW was provided using three strategies. The first approach uses the pixel-level data of frames (UnSup-PLevel). In the other two strategies, a convolutional variational autoencoder (CVAE) is used to extract Un-Supervised Encoded features (UnSup-EnLevel) and Supervised Encoded features (Sup-EnLevel) from the series of Doppler frames. The second approach for equalized data series involves the application of four distinct feature extraction methods: i.e., convolutional neural networks (CNN), supervised and unsupervised CVAE, and principal component Analysis (PCA). The extracted features were considered as an input to the LSTM. This paper presents a comparative analysis of a novel supervised feature extraction pipeline, employing Sup-ENLevel-DTW and Sup-EnLevel-LSTM, against several state-of-the-art unsupervised methods, including UnSUp-EnLevel-DTW, UnSup-EnLevel-LSTM, CNN-LSTM, and PCA-LSTM. The results demonstrate the superiority of the Sup-EnLevel-LSTM strategy. However, the UnSup-PLevel strategy worked surprisingly well without using annotations and frame equalization.

Human activity recognition (HAR) is a crucial task in various applications, including healthcare, fitness, and the military. Deep learning models have revolutionized HAR, however, their computational complexity, particularly those involving BiLSTMs, poses significant challenges for deployment on resource-constrained devices like smartphones. While BiLSTMs effectively capture long-term dependencies by processing inputs bidirectionally, their high parameter count and computational demands hinder practical applications in real-time HAR. This study investigates the approximation of the computationally intensive BiLSTM component in a HAR model by using a combination of alternative model components and data flipping augmentation. The proposed modifications to an existing hybrid model architecture replace the BiLSTM with standard and residual LSTM, along with convolutional networks, supplemented by data flipping augmentation to replicate the context awareness typically provided by BiLSTM networks. The results demonstrate that the residual LSTM (ResLSTM) model achieves superior performance while maintaining a lower computational complexity compared to the traditional BiLSTM model. Specifically, on the UCI-HAR dataset, the ResLSTM model attains an accuracy of 96.34% with 576,702 parameters, outperforming the BiLSTM model’s accuracy of 95.22% with 849,534 parameters. On the WISDM dataset, the ResLSTM achieves an accuracy of 97.20% with 192,238 parameters, compared to the BiLSTM’s 97.23% accuracy with 283,182 parameters, demonstrating a more efficient architecture with minimal performance trade-off. For the KU-HAR dataset, the ResLSTM model achieves an accuracy of 97.05% with 386,038 parameters, showing comparable performance to the BiLSTM model’s 98.63% accuracy with 569,462 parameters, but with significantly fewer parameters.

Human activity recognition (HAR) is pivotal in advancing applications ranging from healthcare monitoring to interactive gaming. Traditional HAR systems, primarily relying on single data sources, face limitations in capturing the full spectrum of human activities. This study introduces a comprehensive approach to HAR by integrating two critical modalities: RGB imaging and advanced pose estimation features. Our methodology leverages the strengths of each modality to overcome the drawbacks of unimodal systems, providing a richer and more accurate representation of activities. We propose a two-stream network that processes skeletal and RGB data in parallel, enhanced by pose estimation techniques for refined feature extraction. The integration of these modalities is facilitated through advanced fusion algorithms, significantly improving recognition accuracy. Extensive experiments conducted on the UTD multimodal human action dataset (UTD MHAD) demonstrate that the proposed approach exceeds the performance of existing state-of-the-art algorithms, yielding improved outcomes. This study not only sets a new benchmark for HAR systems but also highlights the importance of feature engineering in capturing the complexity of human movements and the integration of optimal features. Our findings pave the way for more sophisticated, reliable, and applicable HAR systems in real-world scenarios.

As Canada’s population of older adults rises, the need for aging-in-place solutions is growing due to the declining quality of long-term-care homes and long wait times. While the current standards include questionnaire-based assessments for monitoring activities of daily living (ADLs), there is an urgent need for advanced indoor localization technologies that ensure privacy. This study explores the use of Ultra-Wideband (UWB) technology for activity recognition in a mock condo in the Glenrose Rehabilitation Hospital. UWB systems with built-in Inertial Measurement Unit (IMU) sensors were tested, using anchors set up across the condo and a tag worn by patients. We tested various UWB setups, changed the number of anchors, and varied the tag placement (on the wrist or chest). Wrist-worn tags consistently outperformed chest-worn tags, and the nine-anchor configuration yielded the highest accuracy. Machine learning models were developed to classify activities based on UWB and IMU data. Models that included positional data significantly outperformed those that did not. The Random Forest model with a 4 s data window achieved an accuracy of 94%, compared to 79.2% when positional data were excluded. These findings demonstrate that incorporating positional data with IMU sensors is a promising method for effective remote patient monitoring.

This study explored an indoor system for tracking multiple humans and detecting falls, employing three Millimeter-Wave radars from Texas Instruments. Compared to wearables and camera methods, Millimeter-Wave radar is not plagued by mobility inconveniences, lighting conditions, or privacy issues. We conducted an initial evaluation of radar characteristics, covering aspects such as interference between radars and coverage area. Then, we established a real-time framework to integrate signals received from these radars, allowing us to track the position and body status of human targets non-intrusively. Additionally, we introduced innovative strategies, including dynamic Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering based on signal SNR levels, a probability matrix for enhanced target tracking, target status prediction for fall detection, and a feedback loop for noise reduction. We conducted an extensive evaluation using over 300 min of data, which equated to approximately 360,000 frames. Our prototype system exhibited a remarkable performance, achieving a precision of 98.9% for tracking a single target and 96.5% and 94.0% for tracking two and three targets in human-tracking scenarios, respectively. Moreover, in the field of human fall detection, the system demonstrates a high accuracy rate of 96.3%, underscoring its effectiveness in distinguishing falls from other statuses.