"EWFootstep 1.0"

"Due to the limitations of visual surveillance systems, such as obtrusiveness and high power requirements, audio-based surveillance has gained significant traction in security applications. Among these, footstep-based audio analysis has emerged as a promising and non-intrusive approach for monitoring and threat detection. EWFootstep 1.0, a novel dataset comprising recordings from 176 subjects collected under real-world environmental conditions, distinguishing footstep acoustic signatures of single and multiple persons across forests, roads, and indoor settings. To validate the dataset, we perform time & frequency domain analyses, and implement a CNN-based baseline model. Frechet Audio Distance (FAD) and t-SNE visualizations are carried out to evaluate audio similarity and feature separability across different classes. The dataset bridges the gap in footstep-based security and forensic research by providing a comprehensive dataset for machine learning applications."

由于视觉监控系统存在突兀性强、功耗需求高等局限，基于音频的监控方案在安防应用中获得了广泛青睐。其中，基于脚步声的音频分析作为一种极具前景且无侵入性的监控与威胁检测手段应运而生。本数据集EWFootstep 1.0为全新的脚步声音频数据集，采集了176名受试者的音频样本，录制场景涵盖真实环境下的森林、道路与室内空间，可区分单人与多人行走时的脚步声声学特征。为验证该数据集的有效性，我们开展了时域与频域分析，并搭建了基于卷积神经网络（Convolutional Neural Network, CNN）的基准模型。我们采用弗雷歇音频距离（Frechet Audio Distance, FAD）与t分布邻域嵌入（t-distributed Stochastic Neighbor Embedding, t-SNE）可视化两种方法，评估不同类别音频的相似度与特征可分性。本数据集为基于脚步声的安防与取证研究填补了空白，可为机器学习应用提供完备的数据集支撑。