定义新时代的传感器-based人类活动识别的基石模型：综述与展望
摘要
传感器-based人类活动识别（HAR）支撑着许多普适和可穿戴计算应用，但当前模型受限于标签稀缺、传感器异构性以及在用户、设备和情境之间弱泛化。基石模型通常通过大规模自监督和多模态学习进行预训练，提供一种统一的范式来解决这些挑战，通过学习可复用、可适应的表征来理解活动。本综述综合了面向传感器-based HAR的新兴基石模型。我们首先阐明基本概念、定义和评估标准，然后使用生命周期导向的分类法组织既有工作，涵盖输入设计、预训练、适应和利用。与列举单个模型不同，我们分析跨九个技术轴的重复设计模式和权衡，包括模态范围、标记化、架构、学习范式、适应机制和部署设置。从这些综述中，我们识别出三个主要的发展轨迹：（1）在大型传感器语料库上从头训练的HAR特定基石模型；（2）将通用时间序列或多模态基石模型适应用于传感器-based HAR；（3）集成大型语言模型用于推理、标注和人机交互。我们 conclusions by highlighting open challenges in data curation, multimodal alignment, personalization, privacy, and responsible deployment, and outline directions toward general-purpose, interpretable, and human-centered foundation models for activity understanding. A complete, continuously updated index of papers and models is available in our companion repository: https://github.com/zhaxidele/Foundation-Models-Defining-A-New-Era-In-Human-Activity-Recognition.
引用
@article{arxiv.2604.02711,
title = {Foundation Models Defining A New Era In Sensor-based Human Activity Recognition: A Survey And Outlook},
author = {Sizhen Bian and Mengxi Liu and Lala Shakti Swarup Ray and Bo Zhou and Bin Guo and Zhiwen Yu and Thomas Ploetz and Paul Lukowicz and Siyu Yuan and Vitor Fortes Rey},
journal= {arXiv preprint arXiv:2604.02711},
year = {2026}
}