Self-supervised Learning for Accelerometer-based Human Activity Recognition: A Survey

Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies, Год журнала: 2024, Номер 8(4), С. 1 - 42

Опубликована: Ноя. 21, 2024

Self-supervised learning (SSL) has emerged as a promising alternative to purely supervised learning, since it can learn from labeled and unlabeled data using pre-train-then-fine-tune strategy, achieving state-of-the-art performances across many research areas. The field of accelerometer-based human activity recognition (HAR) benefit SSL be collected cost-efficiently due the ubiquitous nature sensors embedded in smart devices, which is contrast data, that require costly annotation process. Motivated by success lack surveys on for HAR, this survey comprehensively examines 52 methods applied categorizes them into four paradigms based pre-training objectives. We discuss strategies, evaluation protocols, utilized datasets. highlight limitations current methodologies, including little large-scale pre-training, absence foundation models, well scarcity systematic domain shift experiments knowledge utilization. Notably, diversity protocols papers poses considerable challenge when comparing methods. Future directions outlined include development an framework HAR enable standardized benchmarking along with integrating enhance model performance.

Язык: Английский

---

Machine learning models for sequential motion recognition in human activity for logistics operations efficiency improvement

Artificial intelligence for engineering design analysis and manufacturing, Год журнала: 2025, Номер 39

Опубликована: Янв. 1, 2025

Abstract Human activity recognition (HAR) is a vital component of human–robot collaboration. Recognizing the operational elements involved in an operator’s task essential for realizing this vision, and HAR plays key role achieving this. However, recognizing human industrial setting differs from daily living activities. An must be divided into fine to ensure efficient completion. Despite this, there relatively little related research literature. This study aims develop machine learning models classify sequential movement task. To illustrate three logistic operations integrated circuit (IC) design house were studied, with participants wearing 13 inertial measurement units manufactured by XSENS mimic tasks. The kinematics data collected models. time series preprocessing applying two normalization methods different window lengths. Eleven features extracted processed train classification Model validation was carried out using subject-independent method, excluded training dataset. results indicate that developed model can efficiently when operator performs accurately. incorrect classifications occurred missed operation or awkwardly performed RGB video clips helped identify these misclassifications, which used supervisors purposes engineers work improvement.

Язык: Английский

---

A Spatiotemporal Graph Transformer Network for real-time ball trajectory monitoring and prediction in dynamic sports environments

Alexandria Engineering Journal, Год журнала: 2025, Номер 119, С. 246 - 258

Опубликована: Фев. 5, 2025

Язык: Английский

---

Comparison of Deep Learning and Machine Learning Approaches for the Recognition of Dynamic Activities of Daily Living

Lecture notes in computer science, Год журнала: 2025, Номер unknown, С. 18 - 39

Опубликована: Янв. 1, 2025

Язык: Английский

---

Fine-Grained Human Activity Recognition Through Dead-Reckoning and Temporal Convolutional Networks

Lecture notes in computer science, Год журнала: 2025, Номер unknown, С. 3 - 17

Опубликована: Янв. 1, 2025

Язык: Английский

---