Cyber-Physical Anomaly Detection in Smart Homes

Smart homes contain programmable electronic devices (mostly IoT) that enable home automation. People who live in smart homes benefit from interconnected devices by controlling them either remotely or manually/autonomously. However, high interconnectivity comes with an increased attack surface, making the smart home an attractive target for adversaries. NCC Group and the Global Cyber Alliance recorded over 12,000 attacks to log into smart home devices maliciously. Recent statistics show that over 200 million smart homes can be subjected to these attacks. Conventional security systems are either focused on network traffic (e.g., firewalls) or physical environment (e.g., CCTV or basic motion sensors), but not both. A key challenge in developing cyber-physical security systems is the lack of datasets and test beds. For cyber-physical datasets to be meaningful, they need to be collected in real smart home environments. Due to the inherited difficulties and challenges (e.g. effort, costs, test-bed availability), such cyber-physical smart home datasets are quite rare. This paper aims to fill this gap by contributing a dataset we collected in a real smart home with annotated labels. This paper explains the process we followed to collect the data and how we organised them to facilitate wider use within research communities. 

智能家居搭载可编程电子设备（多数为物联网（Internet of Things, IoT）设备），可实现家庭自动化功能。智能家居使用者可通过远程操控、手动操作或自主运行的方式，依托互联设备获取便捷使用体验。然而，高度互联的特性也导致攻击面扩大，使得智能家居成为攻击者极具吸引力的目标。
NCC集团（NCC Group）与全球网络联盟（Global Cyber Alliance）曾记录到超12000次针对智能家居设备的恶意登录攻击。最新统计数据表明，全球已有超2亿智能家居设备面临此类攻击风险。
传统安全系统要么聚焦于网络流量防护（如防火墙），要么仅关注物理环境安全（如闭路电视（Closed-Circuit Television, CCTV）或基础运动传感器），无法兼顾两类防护需求。开发信息物理融合安全系统的核心挑战之一，便是缺乏可用的数据集与测试床（test bed）。若要让信息物理融合数据集具备实际研究价值，需在真实智能家居环境中完成采集。但受限于诸多固有困难与挑战（如人力投入、经费成本、测试床可用性不足等），此类真实场景下的智能家居信息物理融合数据集十分稀缺。
本研究旨在填补这一研究空白，公开我们在真实智能家居环境中采集并完成标注的数据集。本文详细说明了数据采集的完整流程，以及数据的结构化组织方式，以推动该数据集在研究社区中的广泛应用。

搜集汇总

背景与挑战

背景概述

这是一个专注于智能家居环境中网络物理异常检测的数据集，旨在解决现有安全系统通常只关注单一维度（网络或物理）而缺乏融合数据的问题。该数据集在真实智能家居环境中收集，包含标注标签，并涵盖视频、物理传感器和网络流量等多模态数据，规模较大（如网络数据达156.28 GB），适用于物联网安全、人工智能和机器学习等跨学科研究。