Federated learning (FL) enables collaborative model training on decentralized Internet of Things (IoT) data while keeping raw data local, thereby mitigating privacy risks. In practice, however, IoT deployments suffer from two coupled challenges: strict uplink bandwidth constraints and long-tailed label distributions where rare events are most critical. Existing work treats these issues separately—communication-efficient FL often neglects data imbalance, whereas federated long-tail methods typically incur heavy communication overheads and privacy risks. We propose FELT, short forFederated Ensemble Learning for Long-Tailed IoT Data, a framework that jointly addresses communication efficiency, long-tail robustness, and privacy through a communication-efficient private voting protocol. Instead of transmitting full model updates, FELT devices function as teacher models, transmitting only single-integer votes on public queries. This architecture reduces communication traffic by up to one order of magnitude (approx. 10×) compared to standard FL methods. On the server, FELT aggregates votes with calibrated noise to ensure (ε, δ)-differential privacy, followed by a post-hoc class-prior-based calibration that improves tail-class predictions with negligible extra computational overhead on the server side. Experiments on multiple real-world IoT datasets demonstrate that FELT substantially boosts tail-class F1 scores (up to 26% improvement) under the same privacy budget. These results highlight FELT as a practical solution for communication-constrained and privacy-sensitive IoT applications.

Human Activity Recognition (HAR) has emerged as a critical component in intent-aware, AI-driven Internet of Things (IoT) Communication systems, enabling context-aware responses in smart environments. Recently, WiFi-based HAR has gained significant attention due to its non-intrusive nature, low deployment cost, and ability to preserve privacy. However, they face a major challenge across domains. To address this limitation, we propose a novel cross-domain HAR framework (called TransHAR) by introducing a lightweight and efficient transformer model. On one hand, we design a feature representation block that processes the Wi-Fi channel frequency response (CFR) phase data to estimate Doppler shifts, capturing motion-related dynamics while remaining invariant to static, environment-specific structures, enhancing generalization across domains. On the other hand, we propose a lightweight Transformer architecture, termed ResDyTFormer, which minimizes reliance on normalization layers by incorporating a novel Residual Dynamic Tanh function. This function dynamically learns to balance between traditional normalization and the Dynamic Tanh operation, thereby maintaining training stability and avoiding gradient vanishing issues often encountered when using Dynamic Tanh alone. Extensive experiments on two benchmark datasets demonstrate that the proposed TransHAR framework achieves state-of-the-art performance in both in-domain and cross-domain HAR tasks with only 0.17M parameters. On the SHARP dataset, it attains an impressive 99.04% F1 score and 98.94% accuracy. On the 3DO dataset, it achieves 86.10% accuracy and 84.95% F1 score. These results highlight the potential of TransHAR as an efficient and scalable framework for real-world WiFi-based human activity sensing.

Radar-based gesture recognition and person identification offer a natural, convenient, and privacy-preserving approach to human-computer interaction. However, most existing research focuses predominantly on learning for a single task, which requires separate models for each task. This separation increases the complexity of the deployment and the computational overhead. To address these challenges, this study introduces a multi-task learning framework that simultaneously performs gesture recognition and person identification using a miniature radar sensor. By leveraging radar's capacity to capture finegrained spectral and spatial motion patterns, the framework incorporates micro-Doppler and range-Doppler processing, alongside a multi-branch architecture to enhance modality-specific feature representation. It enables unified learning of shared and task-specific features within a single network architecture. The proposed model, MTxLSTM, integrates CNN and the recent xLSTM to mitigate task interference, improve generalization, and improve gesture recognition through person-specific nuances while enhancing person identification by leveraging contextual gesture information. Experimental results reveal that MTxLSTM outperforms existing multi-task learning frameworks and stateof- the-art models, achieving 99.21% in gesture recognition and 98.59% in person identification with moderate model complexity and inference speed. This study concurrently executes gesture recognition and person identification using a miniature radar sensor, and marking the first application of xLSTM in radar sensing technology.

Pedestrian trajectory prediction is a critical task in various mobile computing applications, such as video surveillance, robot navigation, autonomous driving, and human mobility analysis. Although significant progress has been made by current methods, the challenge of observation deficiency in pedestrian trajectory prediction remains largely unaddressed. Since most existing methods focus on optimizing prediction accuracy under the assumption of complete observations, while ignoring the potential for observation deficiency caused by failures in detection or tracking algorithms. To overcome this challenge, we propose a collaborative observation imputation and trajectory prediction framework, which employs consistency evaluation to jointly perform the imputation and prediction tasks. Specifically, we first build a consistency evaluation module to align features between observed and future trajectory pairs using contrastive learning. Then, we design a trajectory imputation and prediction baseline, which adopts a parallel paradigm, to mitigate the impact of coarse imputations on trajectory prediction when performing initial imputation and prediction. Next, we introduce a consistency-guided Skip-Diffusion module, which leverages consistency evaluation between initial imputations and ground truth future trajectories to refine the initial imputations. Finally, we propose a consistency-driven Cross-Mamba module, which uses consistency evaluation between ground-truth observations and initial predictions to refine the initial predictions. Extensive experiments demonstrate the effectiveness of the proposed framework in both imputation and prediction tasks.

Radar-based human activity recognition (HAR) has emerged as a critical component in various applications, ranging from smart homes to healthcare monitoring. Radar has several advantages: a wide detection range, a certain penetration ability, non-contact and non-perception detection ability, not being affected by light, and privacy-preserving. However, achieving high accuracy with efficiency remains a significant challenge due to the complexity of radar signals and the variability in human activities. Currently, the majority of research efforts are centered on enhancing performance, often at the expense of computational efficiency. In this paper, we propose RadarAttn, a novel approach that integrates visual attention mechanisms with self-attention to enhance the performance and efficiency of HAR systems. The architecture of RadarAttn can reduce floating-point operations (FLOPs) and parameter counts while improving accuracy. Our method leverages the visual attention mechanism to focus on the most relevant regions of radar spectrograms. Simultaneously, the self-attention mechanism is used to capture long-range dependencies within the radar signal, enabling the model to learn complex patterns associated with different activities. Experimental results on benchmark radar-based HAR datasets demonstrate that RadarAttn significantly outperforms state-of-the-art methods in both accuracy and computational efficiency. Our approach offers a promising direction for developing robust and scalable radar-based HAR systems for real-world applications.

As the global aging population grows, the need for non-invasive, reliable monitoring solutions for elderly individuals living alone becomes urgent. Kitchen activities, a highrisk area in homes, pose unique safety challenges. However, existing human activity recognition methods still struggle with accuracy, and few studies specifically address these challenges in kitchen environments. This paper introduces OD-GCResNet, a novel hybrid deep learning model for kitchen activity recognition based on micro-Doppler signatures. The proposed model combines Omni-Dimensional Dynamic Convolution with Recursive Gated Convolution to enhance global feature interactions and adaptive attention, allowing for accurate detection of subtle micro-Doppler variations in complex, real-world environments. We validate OD-GCResNet on our collected kitchen activity dataset and achieve a classification accuracy of 99.61 %, outperforming baseline models. This work represents a significant step forward in non-contact, privacy-preserving safety monitoring solutions for elderly care. Our dataset and codes are available at https://github.com/Canberra1111/Kitchen-Micro-Doppler-HAR-with-OD-GCResNet.

With the growing interest in embodied and spatial intelligence, accurately predicting trajectories in 3D environments has become increasingly critical. However, no datasets have been explicitly designed to study 3D trajectory prediction. To this end, we contribute a 3D motion trajectory (3DMoTraj) dataset collected from unmanned underwater vehicles (UUVs) operating in oceanic environments. Mathematically, trajectory prediction becomes significantly more complex when transitioning from 2D to 3D. To tackle this challenge, we analyze the prediction complexity of 3D trajectories and propose a new method consisting of two key components: decoupled trajectory prediction and correlated trajectory refinement. The former decouples inter-axis correlations, thereby reducing prediction complexity and generating coarse predictions. The latter refines the coarse predictions by modeling their inter-axis correlations. Extensive experiments show that our method significantly improves 3D trajectory prediction accuracy and outperforms state-of-the-art methods. Both the 3DMoTraj dataset and the method are available at https://github.com/zhouhao94/3DMoTraj.

Human activity recognition (HAR) based on millimeter wave (mmWave) radar has recently attracted significant interest due to its diverse applications in intelligent robots and human-computer interaction (HCI), including the healthcare monitoring robot. 2-dimensional (2D) histogram features of radar point clouds have demonstrated high accuracy in HAR. But further expansion and refinement of this technique is needed. This paper presents a new precise non-invasive HAR framework based on radar point cloud 2D histograms. Our method enhances conventional 2D histograms by integrating fixed radar sensing boundaries into the histograms, which shows the relative spatial position changes of the target points detected by radar. Additionally, we have concatenated Doppler features (i.e., range-Doppler and angle-Doppler histograms) with the point cloud histograms, resulting in a more comprehensive feature representation than conventional point cloud histograms. We investigated the overfitting issue in stacked hybrid networks and established a multi-layer hybrid network with an optimal number of stacked layers for HAR. In the evaluation, our approach achieves state-of-the-art accuracy, with 99.72% on mmWaveRadarWalking dataset and 98.67% on CI4R-Human-Activity-Recognition dataset, respectively. The proposed method can be applied in the fields of robotics and HCI.

Underwater federated learning (UFL) is an emerging technology to realize distributed intelligent collaboration in the Internet of Underwater Things (IoUT), but its application faces two challenges: the limited bandwidth of underwater communication leads to low model transmission efficiency, and the data is characterized by low quality and high heterogeneity due to environmental interference. In this paper, an underwater federated learning framework with dual-path collaborative optimization is proposed to solve the above problems systematically through the joint design of knowledge distillation and data quality enhancement. Specifically, to optimize the transmission efficiency, a knowledge distillation mechanism is designed, and the complex model is compressed into a simplified model suitable for low-bandwidth transmission by using the collaborative distillation of lightweight teacher-student models. To enhance data quality, a supervised data quality enhancement (S-DQE) method is proposed. The integration of traditional methods with deep learning-based approaches optimizes feature representation through the joint application of contrastive learning and adversarial training, thereby effectively addressing the issue of low-quality underwater data. Finally, numerical results are given to compare the final scheme with the initial federated learning scheme, lightweight model scheme, and lightweight-data quality enhancement scheme, clearly demonstrating its performance gains.

In distributed learning, the End-Edge-Cloud architecture is gaining widespread adoption. However, the cross-layer data interaction in EEC systems significantly increases the risk of privacy breaches. To address this challenge, this paper proposes a novel dynamic privacy-protection framework named MetaDP-HE. The framework is designed to enhance privacy protection while maintaining model performance. It integrates meta-model guided differential privacy (MetaDP) with CKKS homomorphic encryption that supports floating-point operations. A dynamic coordination mechanism is introduced to optimize the parameter configurations between MetaDP and HE. Specifically, clients use the meta-model to predict privacy budgets based on data sensitivity and adjust the noise in differential privacy accordingly. Edge servers then dynamically adjust the encryption parameters of homomorphic encryption based on the noise level, achieving adaptive regulation of encryption strength. The combination of layered encryption and dynamic parameter optimization enables the system to ensure privacy protection and efficient operations when handling data at different levels. Experimental results show that MetaDP-HE outperforms traditional single-privacy methods in both privacy protection and model performance, validating its effectiveness and applicability in practical scenarios.

Metacomputing optimizes distributed computing resources to enhance federated learning systems by enabling efficient resource allocation, improved scheduling, and greater scalability, thereby addressing challenges in large-scale and dynamic environments. This paper proposes an innovative framework integrating Directed Acyclic Graph (DAG) technology with federated learning within a metacomputing environment. The key contributions include a three-layer decentralized federated learning model integrating DAG and metacomputing to enhance resilience and scalability, two advanced tip selection models LazyEval Tip Selector and Precision Tip Selector to optimize node selection and improve data flow, and a Benchmark Improvement Protocol (BIP) for efficient node publishing and role adaptation.The BIP ensures that only high-performing models are published by comparing new models against established benchmarks, which enhances node collaboration and optimizes resource allocation. LazyEval Tip Selector minimizes redundant computations by leveraging a global cache and employing a lazy evaluation strategy, thereby improving computational efficiency. On the other hand, Precision Tip Selector uses a precise scoring mechanism to ensure accurate tip selection, thereby enhancing the robustness and reliability of the entire system. Collectively, these innovations enhance model training efficiency, support real-time updates, and improve the scalability of federated learning systems, making them well-suited for managing complex, dynamic environments.

Radar sensing is gaining increasing attention due to its unique advantages, including being device-free, privacy-preserving, and capable of penetrating obstacles. It has been extensively studied in various applications such as human activity recognition, vital sign monitoring, and person identification. However, most existing research focuses on a single specific application, and there remains a lack of studies or datasets dedicated to multi-task radar sensing. In this paper, we collected a dataset for two sensing tasks, including gesture recognition and person identification, via a miniature mm-wave radar. The raw radar signals were processed using micro-Doppler and range-Doppler techniques to extract spectral and spatial representations. We propose an improved multi-task radar sensing framework (MT-DualFormer) that incorporates attention-based cross-task feature distillation and contrastive learning to maximize task performance. MT-DualFormer consists of dual branches with CNN and Transformer modules, capturing both spatial and temporal dependencies in radar data. Attention-based cross-task feature distillation enables knowledge transfer between gesture recognition and person identification tasks. Meanwhile, contrastive learning ensures embedding space separability, facilitating robust task-specific classification. In the evaluation, MT-DualFormer achieves accuracy rates of 98.87% for gesture recognition and 97.96% for person identification, surpassing five representative multi-task approaches and ten state-of-the-art models. This study underscores the importance of leveraging task correlations to enhance the performance of radar-based sensing systems.

With the continuous development of 6G technology and the Internet of Things, small terminal devices are gradually joining deep model training through wireless networks, leading to the evolution of federated learning. In comparison to traditional centralized learning, federated learning not only leverages the computational power of individual terminals but also ensures the security of terminal data. However, the increasing number of devices poses new requirements on resource utilization in federated learning at scale. In this paper, we aim to address these challenges by proposing an energy-efficient and adaptive resource allocation strategy for wireless heterogeneous layered federated learning model (HLFLM). Specifically, we deploy both macro base stations and multiple micro base stations to construct a HLFLM, and perform resource allocation for subcarriers and power optimization. This approach focuses on optimizing energy consumption in federated learning networks while enhancing scalability and real-time performance of wireless communication. Experimental results demonstrate the effectiveness of the proposed method in medium-sized scenarios.

Current research in human activity recognition primarily emphasizes enhancing accuracy, with limited exploration into computational efficiency and hardware compatibility. Recently, Mamba has sparked substantial interest within the realm of deep learning. Mamba is a hardware-aware algorithm enabling very efficient training and inference. Researchers are applying Mamba to various tasks, demonstrating significant promise in both language and vision tasks. It is worthwhile to investigate the use of Mamba for efficient human activity recognition. In this paper, we proposed a hybrid neural network that integrates CNN and visual Mamba, called ActivityMamba. The SE-Mamba block in ActivityMamba utilizes both CNN’s local and Mamba’s global context modeling while keeping computation and memory efficiency. We evaluated the ActivityMamba on five public benchmark datasets collected by using three different sensing techniques. ActivityMamba achieved higher performance than vision transformers, vision Mamba, and CNNs with fewer FLOPs and parameters. It sets a new SOTA on all five datasets, which are 91.78% OA and 89.13% F1 on the USC-HAD dataset, 99.19% OA and 98.64% F1 on the UT-HAR dataset, 99.82% OA and F1 on the DIAT dataset, 98.59% OA and 98.65% F1 on the UCI-HAR dataset, and 95.41% OA and 93.14% F1 on the UniMib dataset. Our work is the first to investigate the CNN-Mamba hybrid network for efficient human activity recognition.

Transformer architectures are popularized in both vision and natural language processing tasks, and they have achieved new performance benchmarks because of their long-term dependencies modeling, efficient parallel processing, and increased model capacity. While transformers offer powerful capabilities, their demanding computational requirements clash with the real-time and energy-efficient needs of edge-oriented human activity recognition. It is necessary to compress the transformer to reduce its memory consumption and accelerate the inference. In this paper, we investigated the binarization of a transformer-DeepViT for efficient human activity recognition. For feeding sensor signals into DeepViT, we first processed sensor signals to spectrograms by using wavelet transform. Then we applied three methods to binarize DeepViT and evaluated it on three public benchmark datasets for sensor-based human activity recognition. Compared to the full-precision DeepViT, the fully binarized one (Bi-DeepViT) reduced about 96.7% model size and 99% BOPs (Bit Operations) with only a little accuracy compromised. Furthermore, we explored the effects of binarizing various components and latent binarization of DeepViT to understand their impact on the model. We also validated the performance of Bi-DeepViTs on two wireless sensing datasets. The result shows that a certain partial binarization can improve the performance of DeepViT. Our work is the first to apply a binarized transformer in HAR.

Fall detection and recognition play a crucial role in enabling timely medical interventions for people who are at risk of falls, especially among vulnerable populations like older adults and those with mobility limitations. In this article, a cost-effective integrated sensing and communication system, namely, FallDR, is presented for fall detection and recognition using ultrawideband communication. First, we collected the time of flight information of falls (four types) and nonfall events by 10 participants using FallDR. We then proposed a convolutional neural network incorporated with squeeze-and-excitation blocks to detect and recognize falls based on fall trajectories. It proves that the proposed model is accurate, energy-efficient, and lightweight to achieve 100% accuracy in fall detection and recognition. Our proposed solution is proven to be highly robust against environmental changes, such as interference, distance, and direction changes. Further tests in an office showed that FallDR could achieve nearly 100% accuracy, even when the environment was changed. FallDR efficiently employs the characteristics of fall trajectory and the advanced modeling ability of the neural network. We have published our archived data sets and code for comparisons and improvements.

Wireless sensing and communication evolved separately in the past. However, Integrated Sensing and Communication (ISAC) unlocks a new era of mobile network capabilities, with WiFi emerging as a prime candidate. By leveraging existing WiFi infrastructure and frequencies, ISAC enables powerful services like accurate localization and human activity recognition (HAR). WiFi-based HAR is a prime example powered by the magic of ISAC. WiFi Channel State Information (CSI) is susceptible to human movement disturbances; the alterations in CSI mirror the dynamic attributes of human activities. Given the intricate relationship between human activities and CSI, numerous deep learning models have been introduced to enhance HAR accuracy. Recently, transformer-based models have achieved excellent performance in various tasks, including speech recognition, natural language processing, and image classification. This has spurred research into incorporating transformer-based models into WiFi sensing applications. However, their application in WiFi-based HAR remains nascent. Vision transformer is well-suited for analyzing WiFi CSI signals in the form of spectra, such as the Doppler frequency spectrum frequently utilized in related studies, owing to its data structure mimicking that of images. In this study, we explored five widely used Vision Transformer architectures (vanilla ViT, SimpleViT, DeepViT, SwinTransformer, and CaiT) for WiFi CSI-based HAR using two publicly available datasets, UT-HAR and NTU-Fi HAR. Our work aims to assess and compare the performance of diverse ViT architectures for WiFi CSI-based HAR and provide guidelines for WiFi-based HAR modeling and ViT selection, considering accuracy, model size, and computational efficiency.

Human activity recognition (HAR) has become a research hotspot because of its wide range of application prospects. It has higher requirements for real-time and powerefficient processing. However, a large amount of data transfer between sensors and servers, and computation-intensive recognition models hinder the implementation of real-time HAR systems. Recently, edge computing has been proposed to address this challenge by moving computational and data storage resources to the sensors, rather than depending on a centralized server/cloud. In this paper, we investigated binary neural networks for edge intelligence-enabled HAR using radar point cloud. Point cloud can provide 3-dimensional spatial information, which is helpful to improve recognition accuracy. Time-series point cloud also brings challenges, such as larger data volume, 4-dimensional data processing, and more intensive computation. To tackle these challenges, we adopt the 2-dimensional histograms for point cloud multi-view processing and propose the EdgeActNet, a binary neural network for point cloud-based human activity classification on edge devices. In the evaluation, the EdgeActNet achieved the best results with average accuracies of 97.63% on the MMActivity dataset and 95.03% on the point cloud samples of the DGUHA dataset respectively; and saved 16.9× memory consumption and 11.5× inference time compared to its full-precision version. Our work also is the first to apply 2D histogram-based multi-view representation and BNNs for timeseries point cloud classification.

Wearable devices equipped with a variety of sensors facilitate the measurement of physiological and behavioral characteristics. Activity-based person identification is considered an emerging and fast-evolving technology in security and access control fields. Wearables, such as smartphones, Apple Watch, and Google glass can continuously sense and collect activity-related information of users, and activity patterns can be extracted for differentiating different people. Although various human activities have been widely studied, few of them (gaits and keystrokes) have been used for person identification. In this article, we performed person identification using two public benchmark data sets (UCI-HAR and WISDM2019), which are collected from several different activities using multimodal sensors (accelerometer and gyroscope) embedded in wearable devices (smartphone and smartwatch). We implemented eight classifiers, including an multivariate squeeze-and-excitation network (MSENet), time-series transformer (TST), temporal convolutional network (TCN), CNN-LSTM, ConvLSTM, XGBoost, decision tree, and$k$-nearest neighbor. The proposed MSENet can model the relationship between different sensor data. It achieved the best person identification accuracies under different activities of 91.31% and 97.79%, respectively, for the public data sets of UCI-HAR and WISDM2019. We also investigated the effects of sensor modality, human activity, feature fusion, and window size for sensor signal segmentation. Compared to the related work, our approach has achieved the state of the art.

Radar-based human activity recognition is attracting a wide range of interest from both industry and academia because of its through-wall ability, privacy-preserving capability, and device-free detection. Currently, most radar-based systems consider signal analysis and feature extraction in the frequency domain or the temporal domain independently without fusing them together. In this article, in order to model both frequency properties and temporal profiles of human activity, we proposed a spectro-temporal network (STnet) that integrates a temporal convolutional network (TCN) and a convolutional neural network (CNN). It can extract temporal patterns and micro-Doppler features from radar signals for human activity recognition. In the experiments, two radar sensors and one base station were used to build a low-power wireless radar sensor network. Fifteen activities were investigated in a real kitchen scenario by using this radar sensor network. Frequency spectrograms were obtained after signal processing using a short-time Fourier transform (STFT). They were further segmented using a short sliding window (2.5 s), which enables a very small latency. The proposed STnet achieved 99.64% overall accuracy (OA) in testing, which is superior to the other three networks that we implemented in this work. Our work also can be used as a generic solution to other sensor-based (wearable sensors, WiFi channel state information (CSI), etc.) activity recognition.

A wide diversity of sensors has been applied in human activity recognition. These sensors generate enormous amounts of data during human activity monitoring. The long-distance data traveling between sensors and servers increases the costs of bandwidth and latency. However, human activity recognition has a high demand for real-time processing. Recently, edge computing is surging to solve this problem by moving computation and data storage closer to the sensor devices, rather than relying on a central server/cloud. Edge servers are usually designed for low power, low cost, and low computation. They do not support computation-intensive deep learning algorithms or will result in high latency. Fortunately, the development of binarized neural networks enables edge intelligence which supports AI running at the network edge for real-time applications. In this paper, we implement a binarized neural network (BinaryDilatedDenseNet) to enable low-latency and low-memory human activity recognition at the network edge. We applied the BinaryDilatedDenseNet on three sensor-based human activity recognition datasets and evaluated it with four metrics. In comparison, the BinaryDilatedDenseNet outperforms the related work and other three binarized neural networks in accuracy and saves 10 memory and 4.5--8 inference time compared to the FPDilatedDenseNet(the full-precision version of the BinaryDilatedDenseNet).

Automatic fall detection and recognition are challenging problems. In this paper, a novel solution is proposed based on the trajectories of human falls by using the ultra-wideband (UWB) communication system and machine learning methods for fall detection and recognition. Most previous studies of fall detection based on active UWB sensing used electromagnetic signals directly, which may bring problems like radar clutter, signal coupling, multi-path, fading, and interference. Our proposed method only uses human falls trajectories by passive UWB sensing, which achieved fall recognition performance of 93.26% by using the support vector machine with RBF kernel function (SVM-RBF). Compared with previous research, the superiority of this study is that our solution is robust against interference and environmental changes, which means it is reliable for real-world applications. The archived UWB datasets and code have been already published, which may provide the basis for the comparison of techniques and improvements.

In this article, a cost-effective ultrawideband (UWB) communication system for gesture recognition in a smart home environment is proposed, which uses gesture trajectories and a deep learning model. Most previous studies of gesture recognition using the UWB technology used electromagnetic signals directly, which may bring problems, such as radar clutter, signal coupling, multipath, fading, and interference. However, instead of using UWB’s high-frequency pulse signals, the proposed method only uses gesture trajectories by data positioning. To this end, first, a data set of four gesture activities was created. Then, this data set was trained using a convolutional neural network (CNN) integrated with a squeeze-and-excitation (SE) block, namely, the SE-Conv1D model. Finally, the system was prototyped to interact with appliances in practical smart homes. The experimental data was used to demonstrate the superiority of the SE-Conv1D model in comparison with four baselines: 1) support vector machines; 2)$K$-nearest neighbor; 3) random forest; and 4) binarized neural networks. Experimental results show that all collected gesture activities are correctly recognized with an overall accuracy of over 95%, among which the proposed SE-Conv1D model achieves the best accuracy of 99.48%. The proposed system is a complete end-to-end sensing system specifically designed for tracking and recognizing human gestures, which is robust against interference and changes in distance or direction. In addition, the proposed system can tackle the device selection problems for smart homes, which means it is reliable for real-world applications.

Motion trajectories contain rich information about human activities. We propose to use a 2-D LIDAR to perform multiple people activity recognition simultaneously by classifying their trajectories. We clustered raw LIDAR data and classified the clusters into human and nonhuman classes in order to recognize humans in a scenario. For the clusters of humans, we implemented the Kalman filter to track their trajectories which are further segmented and labeled with corresponding activities. We introduced spatial transformation and Gaussian noise for trajectory augmentation in order to overcome the problem of unbalanced classes and boost the performance of human activity recognition (HAR). Finally, we built two neural networks, including a long short-term memory (LSTM) network and a temporal convolutional network (TCN) to classify trajectory samples into 15 activity classes collected from a kitchen. The proposed TCN achieved the best result of 99.49% in overall accuracy. In comparison, the TCN is slightly superior to the LSTM network. Both the TCN and the LSTM network outperform the hidden Markov model (HMM), dynamic time warping (DTW), and support vector machine (SVM) with a wide margin. Our approach achieves a higher activity recognition accuracy than the related work.

Human activity detection plays a crucial role in the recognition of activities of daily living (ADLs). In the past ten years, research on activity detection in the home was achieved through the data aggregation from several different sensors (presence sensors, door contacts, appliances tagging, cameras, wearable beacons, mobile phones, etc.). However, the cost of deployment and maintenance of a multitude of sensor devices and the intrusiveness they can infer are quite high. Research on minimal and non-intrusive sensing for recognition of ADLs are vital for the future of remote care. In this paper, we propose a minimal and non-intrusive low-power low-cost radar-based sensing network system that uses an innovative approach for recognizing human activity in the home. We applied our novel approach to the challenging problem of kitchen activity recognition and investigated fifteen different activities. We designed and trained a deep convolutional neural network (DCNN) that classifies different activities based on their distinct micro-Doppler signatures. We achieved an overall classification rate of 92.8% in activity recognition. Most importantly, in nearly real-time, our approach successfully recognized human activities in more than 89% of the time.