Human Activity Recognition Tasks Research Articles

Modality aware contrastive learning for multimodal human activity recognition

SummaryHuman activity recognition is a well‐established research problem in ubiquitous computing. The increased dependency on various smart devices in our daily lives allows us to investigate the sensor data world produced by multimodal sensors embedded in smart devices. However, the raw sensor data are often unlabeled and annotating this vast amount of data are a costly exercise that can often lead to privacy breaches. Self‐supervised learning‐based approaches are at the forefront of learning semantic representation from unlabeled sensor data, including when applied to human activity recognition tasks. As inferring human activity depends on multimodal sensors, addressing the modality difference and inter‐modality dependencies in a model is an important process. This paper proposes a novel self‐supervised learning approach, modality aware contrastive learning (MACL), for representation learning using multimodal sensor data. The approach uses different sensing modalities to create different views of an input signal. Thus, the model is able to learn the representations by maximizing the similarity among different sensing modalities of the same input signal. Extensive experiments were performed on four publicly available human activity recognition data sets to verify the effectiveness of our proposed MACL method. The experimental evaluation results show that the MACL method attains a comparable performance for human activity recognition to the compared baseline models, directly exceeding the performance of models using standard augmentation transformation strategies.

A micro-Doppler spectrogram denoising algorithm for radar human activity recognition

Radar signal recognition based on micro-Doppler spectrogram has been widely used in human action recognition tasks. However, in practical application scenarios, radar signals inevitably have noise, which leads to different degrees of deformation of the spectrogram graph structure and affects the accuracy of subsequent recognition algorithms. In this paper, we present “ACFL”, a novel algorithm for micro-Doppler spectrogram denoising, which aims to reduce the impact of noise on human action recognition. ACFL employs amplitude–frequency two-dimensional clustering and fuzzy logic clustering selection mechanism to remove noise elements from the spectrogram. Moreover, to address the issue of noise leakage or target missing under time-varying noise and action conditions, ACFL adopts spectrogram segmentation based on short-term Rényi entropy. By dividing the spectrogram into intervals with different time–frequency distributions, the dynamic spectrogram denoise over time is achieved. Simulation and measured data experiments demonstrate that the proposed algorithm not only achieves a higher-quality denoised spectrogram but also significantly improves the accuracy of human action recognition under noisy conditions.

Spatio-Temporal Fusion for Human Action Recognition via Joint Trajectory Graph

Graph Convolutional Networks (GCNs) and Transformers have been widely applied to skeleton-based human action recognition, with each offering unique advantages in capturing spatial relationships and long-range dependencies. However, for most GCN methods, the construction of topological structures relies solely on the spatial information of human joints, limiting their ability to directly capture richer spatio-temporal dependencies. Additionally, the self-attention modules of many Transformer methods lack topological structure information, restricting the robustness and generalization of the models. To address these issues, we propose a Joint Trajectory Graph (JTG) that integrates spatio-temporal information into a uniform graph structure. We also present a Joint Trajectory GraphFormer (JT-GraphFormer), which directly captures the spatio-temporal relationships among all joint trajectories for human action recognition. To better integrate topological information into spatio-temporal relationships, we introduce a Spatio-Temporal Dijkstra Attention (STDA) mechanism to calculate relationship scores for all the joints in JTG. Furthermore, we incorporate the Koopman operator into the classification stage to enhance the model's representation ability and classification performance. Experiments demonstrate that JT-GraphFormer achieves outstanding performance in human action recognition tasks, outperforming state-of-the-art methods on the NTU RGB+D, NTU RGB+D 120, and N-UCLA datasets.

MFCANN: A feature diversification framework based on local and global attention for human activity recognition

Human activity recognition (HAR) is a crucial detection technique extensively employed in various contexts demanding accurate identification of human actions. Currently, mainstream HAR approaches frequently rely on sequential data generated by wearable sensors, with a primary focus on extracting feature representations that align with various human activities. Nevertheless, complex and diverse human activities encompass various kinds of features and substantial variations in the features emphasized across different activities, making it challenging for current artificial intelligence methods to achieve accurate recognition. To address this challenge, this paper introduces a network framework called the Multi-Feature Combining Attention Neural Network (MFCANN). This framework incorporates diverse feature extraction, along with both local and global feature attention mechanisms. It is composed of stacked Multi-Feature Combining Attention Blocks (MFCAB) designed by us. In contrast to traditional convolutional methods, MFCAB parallelly stacks various convolutional components, enabling the extraction of a broader range of features from human activity data. Additionally, we propose the Intra-Module Attention Block (Intra-MAB) and the Inter-Module Attention Block (Inter-MAB), which simultaneously focus on local fine-grained features within module feature maps and global distinguishing features across module feature maps. This aims to achieve more targeted feature learning, reinforcing the network’s ability to distinguish between different human activities. Abundant experimental results demonstrate that the proposed MFCANN in this paper outperforms current mainstream deep learning algorithms in HAR tasks, achieving recognition accuracies of 0.9813, 0.9324, and 0.9930 on the UCI-HAR, USC-HAD, and RealWorld datasets respectively.

A Systematic Evaluation of Recurrent Neural Network Models for Edge Intelligence and Human Activity Recognition Applications

The Recurrent Neural Networks (RNNs) are an essential class of supervised learning algorithms. Complex tasks like speech recognition, machine translation, sentiment classification, weather prediction, etc., are now performed by well-trained RNNs. Local or cloud-based GPU machines are used to train them. However, inference is now shifting to miniature, mobile, IoT devices and even micro-controllers. Due to their colossal memory and computing requirements, mapping RNNs directly onto resource-constrained platforms is arcane and challenging. The efficacy of edge-intelligent RNNs (EI-RNNs) must satisfy both performance and memory-fitting requirements at the same time without compromising one for the other. This study’s aim was to provide an empirical evaluation and optimization of historic as well as recent RNN architectures for high-performance and low-memory footprint goals. We focused on Human Activity Recognition (HAR) tasks based on wearable sensor data for embedded healthcare applications. We evaluated and optimized six different recurrent units, namely Vanilla RNNs, Long Short-Term Memory (LSTM) units, Gated Recurrent Units (GRUs), Fast Gated Recurrent Neural Networks (FGRNNs), Fast Recurrent Neural Networks (FRNNs), and Unitary Gated Recurrent Neural Networks (UGRNNs) on eight publicly available time-series HAR datasets. We used the hold-out and cross-validation protocols for training the RNNs. We used low-rank parameterization, iterative hard thresholding, and spare retraining compression for RNNs. We found that efficient training (i.e., dataset handling and preprocessing procedures, hyperparameter tuning, and so on, and suitable compression methods (like low-rank parameterization and iterative pruning) are critical in optimizing RNNs for performance and memory efficiency. We implemented the inference of the optimized models on Raspberry Pi.

Spatio-temporal neural network with handcrafted features for skeleton-based action recognition

The task of human action recognition (HAR) can be found in many computer vision practical applications. Various data modalities have been considered for solving this task, including joint-based skeletal representations which are suitable for real-time applications on platforms with limited computational resources. We propose a spatio-temporal neural network that uses handcrafted geometric features to classify human actions from video data. The proposed deep neural network architecture combines graph convolutional and temporal convolutional layers. The experiments performed on public HAR datasets show that our model obtains results similar to other state-of-the-art methods but has a lower inference time while offering the possibility to obtain an explanation for the classified action.

Towards Learning Discrete Representations via Self-Supervision for Wearables-Based Human Activity Recognition.

Human activity recognition (HAR) in wearable and ubiquitous computing typically involves translating sensor readings into feature representations, either derived through dedicated pre-processing procedures or integrated into end-to-end learning approaches. Independent of their origin, for the vast majority of contemporary HAR methods and applications, those feature representations are typically continuous in nature. That has not always been the case. In the early days of HAR, discretization approaches had been explored-primarily motivated by the desire to minimize computational requirements on HAR, but also with a view on applications beyond mere activity classification, such as, for example, activity discovery, fingerprinting, or large-scale search. Those traditional discretization approaches, however, suffer from substantial loss in precision and resolution in the resulting data representations with detrimental effects on downstream analysis tasks. Times have changed, and in this paper, we propose a return to discretized representations. We adopt and apply recent advancements in vector quantization (VQ) to wearables applications, which enables us to directly learn a mapping between short spans of sensor data and a codebook of vectors, where the index comprises the discrete representation, resulting in recognition performance that is at least on par with their contemporary, continuous counterparts-often surpassing them. Therefore, this work presents a proof of concept for demonstrating how effective discrete representations can be derived, enabling applications beyond mere activity classification but also opening up the field to advanced tools for the analysis of symbolic sequences, as they are known, for example, from domains such as natural language processing. Based on an extensive experimental evaluation of a suite of wearable-based benchmark HAR tasks, we demonstrate the potential of our learned discretization scheme and discuss how discretized sensor data analysis can lead to substantial changes in HAR.

Enhancing Human Action Recognition with 3D Skeleton Data: A Comprehensive Study of Deep Learning and Data Augmentation

Human Action Recognition (HAR) is an important field that identifies human behavior through sensor data. Three-dimensional human skeleton data extracted from the Kinect depth sensor have emerged as a powerful alternative to mitigate the effects of lighting and occlusion of traditional 2D RGB or grayscale image-based HAR. Data augmentation is a key technique to enhance model generalization and robustness in deep learning while suppressing overfitting to training data. In this paper, we conduct a comprehensive study of various data augmentation techniques specific to skeletal data, which aim to improve the accuracy of deep learning models. These augmentation methods include spatial augmentation, which generates augmented samples from the original 3D skeleton sequence, and temporal augmentation, which is designed to capture subtle temporal changes in motion. The evaluation covers two publicly available datasets and a proprietary dataset and employs three neural network models. The results highlight the impact of temporal augmentation on model performance on the skeleton datasets, while exhibiting the nuanced impact of spatial augmentation. The findings underscore the importance of tailoring augmentation strategies to specific dataset characteristics and actions, providing novel perspectives for model selection in skeleton-based human action recognition tasks.

Domain Adaptation for Sensor-Based Human Activity Recognition with a Graph Convolutional Network

Sensor-based human activity recognition (HAR) plays a fundamental role in various mobile application scenarios, but the model performance of HAR heavily relies on the richness of the dataset and the completeness of data annotation. To address the shortage of comprehensive activity types in collected datasets, we adopt the domain adaptation technique with a graph neural network-based approach by incorporating an adaptive learning mechanism to enhance the action recognition model’s generalization ability, especially when faced with limited sample sizes. To evaluate the effectiveness of our proposed approach, we conducted experiments using three well-known datasets: MHealth, PAMAP2, and TNDA. The experimental results demonstrate the efficacy of our approach in sensor-based HAR tasks, achieving impressive average accuracies of 98.88%, 98.58%, and 97.78% based on the respective datasets. Furthermore, we conducted transfer learning experiments to address the domain adaptation problem. These experiments revealed that our proposed model exhibits exceptional transferability and distinguishing ability, even in scenarios with limited available samples. Thus, our approach offers a practical and viable solution for sensor-based HAR tasks.

Towards Flexible and Low-Power Wireless Smart Sensors: Reconfigurable Analog-to-Feature Conversion for Healthcare Applications.

Analog-to-feature (A2F) conversion based on non-uniform wavelet sampling (NUWS) has demonstrated the ability to reduce energy consumption in wireless sensors while employed for electrocardiogram (ECG) anomaly detection. The technique involves extracting only relevant features for a given task directly from analog signals and conducting classification in the digital domain. Building on this approach, we extended the application of the proposed generic A2F converter to address a human activity recognition (HAR) task. The performed simulations include the training and evaluation of neural network (NN) classifiers built for each application. The corresponding results enabled the definition of valuable features and the hardware specifications for the ongoing complete circuit design. One of the principal elements constituting the developed converter, the integrator brought from the state-of-the-art design, was modified and simulated at the circuit level to meet our requirements. The revised value of its power consumption served to estimate the energy spent by the communication chain with the A2F converter. It consumes at least 20 and 5 times less than the chain employing the Nyquist approach in arrhythmia detection and HAR tasks, respectively. This fact highlights the potential of A2F conversion with NUWS in achieving flexible and energy-efficient sensor systems for diverse applications.

MHAiR: A Dataset of Audio-Image Representations for Multimodal Human Actions

Audio-image representations for a multimodal human action (MHAiR) dataset contains six different image representations of the audio signals that capture the temporal dynamics of the actions in a very compact and informative way. The dataset was extracted from the audio recordings which were captured from an existing video dataset, i.e., UCF101. Each data sample captured a duration of approximately 10 s long, and the overall dataset was split into 4893 training samples and 1944 testing samples. The resulting feature sequences were then converted into images, which can be used for human action recognition and other related tasks. These images can be used as a benchmark dataset for evaluating the performance of machine learning models for human action recognition and related tasks. These audio-image representations could be suitable for a wide range of applications, such as surveillance, healthcare monitoring, and robotics. The dataset can also be used for transfer learning, where pre-trained models can be fine-tuned on a specific task using specific audio images. Thus, this dataset can facilitate the development of new techniques and approaches for improving the accuracy of human action-related tasks and also serve as a standard benchmark for testing the performance of different machine learning models and algorithms.

IMU-Based Fitness Activity Recognition Using CNNs for Time Series Classification.

Mobile fitness applications provide the opportunity to show users real-time feedback on their current fitness activity. For such applications, it is essential to accurately track the user's current fitness activity using available mobile sensors, such as inertial measurement units (IMUs). Convolutional neural networks (CNNs) have been shown to produce strong results in different time series classification tasks, including the recognition of daily living activities. However, fitness activities can present unique challenges to the human activity recognition task (HAR), including greater similarity between individual activities and fewer available data for model training. In this paper, we evaluate the applicability of CNNs to the fitness activity recognition task (FAR) using IMU data and determine the impact of input data size and sensor count on performance. For this purpose, we adapted three existing CNN architectures to the FAR task and designed a fourth CNN variant, which we call the scaling fully convolutional network (Scaling-FCN). We designed a preprocessing pipeline and recorded a running exercise data set with 20 participants, in which we evaluated the respective recognition performances of the four networks, comparing them with three traditional machine learning (ML) methods commonly used in HAR. Although CNN architectures achieve at least 94% test accuracy in all scenarios, two traditional ML architectures surpass them in the default scenario, with support vector machines (SVMs) achieving 99.00 ± 0.34% test accuracy. The removal of all sensors except one foot sensor reduced the performance of traditional ML architectures but improved the performance of CNN architectures on our data set, with our Scaling-FCN reaching the highest accuracy of 99.86 ± 0.11% on the test set. Our results suggest that CNNs are generally well suited for fitness activity recognition, and noticeable performance improvements can be achieved if sensors are dropped selectively, although traditional ML architectures can still compete with or even surpass CNNs when favorable input data are utilized.

Robust Feature Representation Using Multi-Task Learning for Human Activity Recognition.

Learning underlying patterns from sensory data is crucial in the Human Activity Recognition (HAR) task to avoid poor generalization when coping with unseen data. A key solution to such an issue is representation learning, which becomes essential when input signals contain activities with similar patterns or when patterns generated by different subjects for the same activity vary. To address these issues, we seek a solution to increase generalization by learning the underlying factors of each sensor signal. We develop a novel multi-channel asymmetric auto-encoder to recreate input signals precisely and extract indicative unsupervised futures. Further, we investigate the role of various activation functions in signal reconstruction to ensure the model preserves the patterns of each activity in the output. Our main contribution is that we propose a multi-task learning model to enhance representation learning through shared layers between signal reconstruction and the HAR task to improve the robustness of the model in coping with users not included in the training phase. The proposed model learns shared features between different tasks that are indeed the underlying factors of each input signal. We validate our multi-task learning model using several publicly available HAR datasets, UCI-HAR, MHealth, PAMAP2, and USC-HAD, and an in-house alpine skiing dataset collected in the wild, where our model achieved 99%, 99%, 95%, 88%, and 92% accuracy. Our proposed method shows consistent performance and good generalization on all the datasets compared to the state of the art.

CROMOSim: A Deep Learning-Based Cross-Modality Inertial Measurement Simulator

With the prevalence of wearable devices, inertial measurement unit (IMU) data has been utilized in monitoring and assessing human mobility such as human activity recognition (HAR) and human pose estimation (HPE). Training deep neural network (DNN) models for these tasks require a large amount of labelled data, which are hard to acquire in uncontrolled environments. To mitigate the data scarcity problem, we design CROMOSim, a cross-modality sensor simulator that simulates high fidelity virtual IMU sensor data from motion capture systems or monocular RGB cameras. It utilizes a skinned multi-person linear model (SMPL) for 3D body pose and shape representations to enable simulation from arbitrary on-body positions. Then a DNN model is trained to learn the functional mapping from imperfect trajectory estimations in a 3D SMPL body tri-mesh due to measurement noise, calibration errors, occlusion and other modelling artifacts, to IMU data. We evaluate the fidelity of CROMOSim simulated data and its utility in data augmentation on various HAR and HPE datasets. Extensive empirical results show that the proposed model achieves a 6.7% improvement over baseline methods in a HAR task.

Clinical human activity recognition based on a wearable patch of combined tri-axial ACC and ECG sensors.

In digital medicine, human activity recognition (HAR) can be used to track and assess a patient's progress throughout rehabilitation, enhancing the quality of life for the elderly and the disabled. A patch-type flexible sensor that integrated dynamic electrocardiogram (ECG) and acceleration signal (ACC) was used to record the signals of the various behavioral activities of 20 healthy volunteers and 25 patients with pneumoconiosis. Seven HAR tasks were then carried out on the data using four different deep learning methods (CNN, LSTM, CNN-LSTM and GRU). When ECG and ACC were obtained simultaneously, the overall accuracy rates of HAR for healthy group were 0.9371, 0.8829, 0.9843 and 0.9486 by the CNN, LSTM, CNN-LSTM and GRU models, respectively. In contrast, the overall accuracy rates of HAR for the pneumoconiosis patients' group were 0.8850, 0.7975, 0.9425 and 0.8525 by the four corresponding models. The accuracy of HAR for both groups using all four models is higher than when only ACC signal is detected. The addition of the ECG signal significantly improves HAR outcomes in the group of healthy individuals, while having relatively less enhancing effects on the group of patients with pneumoconiosis. When ECG and ACC signals were combined, the increase in HAR accuracy was notable compared to cases where no ECG data was provided. These results suggest that the combination of ACC and ECG data can represent a novel method for the clinical application of HAR.

GLULA: Linear attention-based model for efficient human activity recognition from wearable sensors.

Body-worn sensor data is used in monitoring patient activity during rehabilitation and also can be extended to controlling rehabilitation devices based on the activity of the person. The primary focus of research has been on effectively capturing the spatiotemporal dependencies in the data collected by these sensors and efficiently classifying human activities. With the increasing complexity and size of models, there is a growing emphasis on optimizing their efficiency in terms of memory usage and inference time for real-time usage and mobile computers. While hybrid models combining convolutional and recurrent neural networks have shown strong performance compared to traditional approaches, self-attention-based networks have demonstrated even superior results. However, instead of relying on the same transformer architecture, there is an opportunity to develop a novel framework that incorporates recent advancements to enhance speed and memory efficiency, specifically tailored for human activity recognition (HAR) tasks. In line with this approach, we present GLULA, a unique architecture for HAR. GLULA combines gated convolutional networks, branched convolutions, and linear self-attention to achieve efficient and powerful solutions. To enhance the performance of our proposed architecture, we employed manifold mixup as an augmentation variant which proved beneficial in limited data settings. Extensive experiments were conducted on five benchmark datasets: PAMAP2, SKODA, OPPORTUNITY, DAPHNET, and USC-HAD. Our findings demonstrate that GLULA outperforms recent models in the literature on the latter four datasets but also exhibits the lowest parameter count and close to the fastest inference time among state-of-the-art models.

LHAR: Lightweight Human Activity Recognition on Knowledge Distillation.

Sensor-based Human Activity Recognition (HAR) is widely used in daily life and is the basic-level bridge to virtual healthcare in the metaverse. The current challenge is the low recognition accuracy for personalized users on smart wearable devices. The limited resource cannot support large deep learning models updated locally. Besides, integrating and transmitting sensor data to the cloud would reduce the efficiency. Considering the tradeoff between performance and complexity, we propose a Lightweight Human Activity Recognition (LHAR) framework. In LHAR, we combine the cross-people HAR task with the lightweight model task. LHAR framework is designed on the teacher-student architecture and the student network consists of multiple depthwise separable convolution layers to achieve fewer parameters. The dark knowledge distilled from the complex teacher model enhances the generalization ability of LHAR. To achieve effective knowledge distillation, we propose two optimization methods. Firstly, we train the teacher model by ensemble learning to promote teacher performance. Secondly, a multi-channel data augmentation method is proposed for the diversity of the dataset, which is a plug-in operation for the ensemble teacher model. In the experiments, we compare LHAR with state-of-art models in comparison evaluation, ablation study and the hyperparameter analysis, which proves the better performance of LHAR in efficiency and effectiveness.

An improved human activity recognition technique based on convolutional neural network

A convolutional neural network (CNN) is an important and widely utilized part of the artificial neural network (ANN) for computer vision, mostly used in the pattern recognition system. The most important applications of CNN are medical image analysis, image classification, object recognition from videos, recommender systems, financial time series analysis, natural language processing, and human–computer interfaces. However, after the technological advancement in the power of computing ability and the emergence of huge quantities of labeled data provided through enhanced algorithms, nowadays, CNN is widely used in almost every area of study. One of the main uses of wearable technology and CNN within medical surveillance is human activity recognition (HAR), which must require constant tracking of everyday activities. This paper provides a comprehensive study of the application of CNNs in the classification of HAR tasks. We describe their enhancement, from their antecedents up to the current state-of-the-art systems of deep learning (DL). We have provided a comprehensive working principle of CNN for HAR tasks, and a CNN-based model is presented to perform the classification of human activities. The proposed technique interprets data from sensor sequences of inputs by using a multi-layered CNN that gathers temporal and spatial data related to human activities. The publicly available WISDM dataset for HAR has been used to perform this study. This proposed study uses the two-dimensional CNN approach to make a model for the classification of different human activities. A recent version of Python software has been used to perform the study. The rate of accuracy for HAR through the proposed model in this experiment is 97.20%, which is better than the previously estimated state-of-the-art technique. The findings of the study imply that using DL methods for activity recognition might greatly increase accuracy and increase the range of applications where HAR can be used successfully. We have also described the future research trends in the field of HAR in this article.

Deep Wavelet Convolutional Neural Networks for Multimodal Human Activity Recognition Using Wearable Inertial Sensors.

Recent advances in wearable systems have made inertial sensors, such as accelerometers and gyroscopes, compact, lightweight, multimodal, low-cost, and highly accurate. Wearable inertial sensor-based multimodal human activity recognition (HAR) methods utilize the rich sensing data from embedded multimodal sensors to infer human activities. However, existing HAR approaches either rely on domain knowledge or fail to address the time-frequency dependencies of multimodal sensor signals. In this paper, we propose a novel method called deep wavelet convolutional neural networks (DWCNN) designed to learn features from the time-frequency domain and improve accuracy for multimodal HAR. DWCNN introduces a framework that combines continuous wavelet transforms (CWT) with enhanced deep convolutional neural networks (DCNN) to capture the dependencies of sensing signals in the time-frequency domain, thereby enhancing the feature representation ability for multiple wearable inertial sensor-based HAR tasks. Within the CWT, we further propose an algorithm to estimate the wavelet scale parameter. This helps enhance the performance of CWT when computing the time-frequency representation of the input signals. The output of the CWT then serves as input for the proposed DCNN, which consists of residual blocks for extracting features from different modalities and attention blocks for fusing these features of multimodal signals. We conducted extensive experiments on five benchmark HAR datasets: WISDM, UCI-HAR, Heterogeneous, PAMAP2, and UniMiB SHAR. The experimental results demonstrate the superior performance of the proposed model over existing competitors.

A Lightweight Graph Neural Network Algorithm for Action Recognition Based on Self-Distillation

Recognizing human actions can help in numerous ways, such as health monitoring, intelligent surveillance, virtual reality and human–computer interaction. A quick and accurate detection algorithm is required for daily real-time detection. This paper first proposes to generate a lightweight graph neural network by self-distillation for human action recognition tasks. The lightweight graph neural network was evaluated on the NTU-RGB+D dataset. The results demonstrate that, with competitive accuracy, the heavyweight graph neural network can be compressed by up to 80%. Furthermore, the learned representations have denser clusters, estimated by the Davies–Bouldin index, the Dunn index and silhouette coefficients. The ideal input data and algorithm capacity are also discussed.