Pedestrian Navigation Activity Recognition Based On Segmentation Transformer

IEEE Internet of Things Journal, Journal Year: 2024, Volume and Issue: 11(15), P. 26020 - 26032

Published: May 2, 2024

In the context of Internet Things, utilizing inherent inertial sensors in smartphones for human activity recognition (HAR) has garnered considerable attention owing to its wide-ranging applications. However, prevailing HAR approaches primarily treat identification as a single-label classification task, focusing solely on discerning pedestrian motion modes or device usage modes, while disregarding their interrelatedness. Additionally, methods employing sliding windows encounter challenges associated with multiclass window problem, wherein certain sample labels differ from label assigned window. This paper aims address these issues. presents novel approach simultaneously recognizing and by segmentation transformer. The proposed joint framework effectively annotates sensor data at each timestamp achieves dense prediction time-series through encoding decoding annotated data. To optimize utilization information extracted Transformer layer, global up-sampling decoder based pyramid module is introduced, enabling features obtained layer. We performed experiments two publicly available datasets comprehensively assess effectiveness methodology. results demonstrate that our an accuracy 99.79% weighted F-score 99.77%, surpassing performance existing state-of-the-art methods. Furthermore, we constructed heterogeneous validate robustness method. extensive experimental findings indicate uncovers correlations between leading enhanced addressing posed problem.

Language: Английский

---

Enhancing human activity recognition for the elderly and individuals with disabilities through optimized Internet-of-Things and artificial intelligence integration with advanced neural networks

Frontiers in Neuroinformatics, Journal Year: 2024, Volume and Issue: 18

Published: Nov. 19, 2024

Elderly and individuals with disabilities can greatly benefit from human activity recognition (HAR) systems, which have recently advanced significantly due to the integration of Internet Things (IoT) artificial intelligence (AI). The blending IoT AI methodologies into HAR systems has potential enable these populations lead more autonomous comfortable lives. are equipped various sensors, including motion capture microcontrollers, transceivers, supply data assorted machine learning (ML) algorithms for subsequent analyses. Despite substantial advantages this integration, current frameworks encounter significant challenges related computational overhead, arises complexity ML algorithms. This article introduces a novel ensemble gated recurrent networks (GRN) deep extreme feedforward neural (DEFNN), hyperparameters optimized through water drop optimization (AWDO) algorithm. framework leverages GRN effective feature extraction, subsequently utilized by DEFNN accurately classifying data. Additionally, AWDO is employed within adjust hyperparameters, thereby mitigating overhead enhancing detection efficiency. Extensive experiments were conducted verify proposed methodology using real-time datasets gathered testbeds, employ NodeMCU units interfaced Wi-Fi transceivers. framework's efficiency was assessed several metrics: accuracy at 99.5%, precision 98%, recall 97%, specificity F1-score 98.2%. These results then benchmarked against other contemporary (DL)-based systems. experimental outcomes indicate that our model achieves near-perfect accuracy, surpassing alternative learning-based Moreover, demonstrates reduced demands compared preceding algorithms, suggesting may offer superior efficacy compatibility deployment in designed elderly or disabilities.

Language: Английский

---

Editorial: Second Quarter 2024 IEEE Communications Surveys and Tutorials

IEEE Communications Surveys & Tutorials, Journal Year: 2024, Volume and Issue: 26(2), P. i - vi

Published: Jan. 1, 2024

I welcome you to the second issue of IEEE Communications Surveys and Tutorials in 2024. This includes 18 articles covering different aspects communication networks. In particular, these survey tutor various issues "Wireless Communications", "Cyber Security", "IoT M2M", "Internet Technologies", "Network Virtualization" "Optical Communications". A brief account for each is given below.

Language: Английский

---

UMAHand: A dataset of inertial signals of typical hand activities

Data in Brief, Journal Year: 2024, Volume and Issue: 55, P. 110731 - 110731

Published: July 11, 2024

Given the popularity of wrist-worn devices, particularly smartwatches, identification manual movement patterns has become utmost interest within research field Human Activity Recognition (HAR) systems. In this context, by leveraging numerous sensors natively embedded in HAR functionalities that can be implemented a watch via software and very cost-efficient way cover wide variety applications, ranging from fitness trackers to gesture detectors aimed at disabled individuals (e.g., for sending alarms), promoting behavioral activation or healthy lifestyle habits. regard, development artificial intelligence algorithms capable effectively discriminating these activities, it is great importance have repositories movements allow scientific community train, evaluate, benchmark new proposals detectors. The UMAHand dataset offers collection files containing signals captured Shimmer 3 sensor node, which includes an accelerometer, gyroscope, magnetometer barometer, during execution different typical hand movements. For purpose, measurements four sensors, gathered sampling rate 100 Hz, were taken group 25 volunteers (16 females 9 males), aged between 18 56, performance 29 daily life activities involving mobility. Participants wore node on their dominant throughout experiments.

Language: Английский

---

Human Action Recognition Network Containing Hands Based on NPoseC3D59

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 30, 2024

\abstract{With the intelligent development of machinery manufacturing, importance human-computer interaction has become more prominent, and recognizing human actions is a prerequisite for realizing intelligence interaction. Human limbs subtle hand movements contain rich communication information, combining two in research can accurately capture behavior understand intentions. PoseC3D-based action recognition models are favored their excellent performance low parameter count be friendly to add key point model on top only body torso. Therefore, this study, NPoseC3D constructed dataset containing built effect validation. Firstly, improve accuracy, ReLU activation function layer, BN downsampling method original SlowOnly backbone adjusted retain active trained NTU RGB+D 60 publicly available validation, accuracy improved. Secondly, because there no hands, Whole59, an gestures industrial scenarios, labeled, 42 keypoint inputs added model, tested Whole59 dataset, obtain NPoseC3D59 hand, experimental results prove effectiveness proposed research.

Language: Английский

---

FedMMA-HAR: Federated Learning for Human Activity Recognition With Missing Modalities Using Head-Worn Wearables

IEEE Pervasive Computing, Journal Year: 2024, Volume and Issue: 23(4), P. 40 - 49

Published: Oct. 1, 2024

Language: Английский

---

Song Gesture Recognition for a Robot-Enhanced Imitation Therapy

Published: Aug. 26, 2024

Language: Английский

---

Investigation of Environment Dependence in Wi-Fi CSI-Based Crowd Counting Systems

TENCON 2021 - 2021 IEEE Region 10 Conference (TENCON), Journal Year: 2024, Volume and Issue: unknown, P. 362 - 365

Published: Dec. 1, 2024

Language: Английский

---