Fall Vision: A Benchmark Video Dataset for Advancing Fall Detection Technology

In this paper, an exhaustive video dataset categorized as fall and no-fall videos is presented, which was compiled for the specific purpose of fall detection research. The dataset comprises three fundamental classifications of falls, namely those originating from a standing position, bed, or chair. After being initially acquired in unprocessed form, these videos underwent subsequent processing to generate seminal videos, which were presented with and without a black backdrop. The dataset was obtained from voluntary participants through the use of handheld devices (e.g., digital cameras or mobile phones), which ensured ethical compliance and informed assent. The dataset provides a substantial asset for the progression of fall detection algorithms, serving as a resilient framework for the development and evaluation of such algorithms. The implementation of fall detection systems is critical, especially in situations involving elderly individuals who occur during medical emergencies that lead to falls and require immediate assistance, or when individuals are solitary and unable to restore their balance after falling. By utilizing this dataset, scientists have the opportunity to investigate a wide range of methodologies, such as deep learning and computer vision, in order to develop and enhance fall detection systems. This video dataset has the potential to contribute to the development of fall detection technology, thereby improving safety protocols for vulnerable populations, due to its availability to researchers.

本文提出了一个全面的视频数据集，分为跌倒与非跌倒视频两类，专为跌倒检测（fall detection）研究编制。该数据集包含三大类跌倒场景，分别为从站立位、床或椅子上发生的跌倒。这些视频最初以未处理的原始格式采集，随后经过加工生成核心样本视频，分别提供带黑色背景与不带黑色背景两个版本。本数据集通过手持设备（如数码相机或手机）采集自自愿参与者，确保了研究的伦理合规性并获得了所有参与者的知情同意。该数据集为跌倒检测算法的迭代优化提供了重要资源，可作为此类算法开发与评估的稳健基准框架。跌倒检测系统的部署应用至关重要，尤其适用于以下场景：老年群体因医疗急症发生跌倒且需立即施救时，或是独居者跌倒后无法自行恢复平衡的情况。借助该数据集，研究人员可探索包括深度学习与计算机视觉在内的多种技术方法，以开发并优化跌倒检测系统。由于该数据集可供研究人员使用，其有望推动跌倒检测技术的发展，进而提升脆弱群体的安全保障规范。