Advanced Technology Development Centre, Indian Institute of Technology, Kharagpur, West Bengal, India.
Department of Electrical Engineering, Indian Institute of Technology, Kharagpur, West Bengal, India.
Comput Biol Med. 2023 Sep;164:107315. doi: 10.1016/j.compbiomed.2023.107315. Epub 2023 Aug 8.
Existing low-cost Doppler radar-based fall detection systems encounter challenges due to false alarms and the absence of post-fall health tracking, significantly impacting their accuracy and overall compatibility for fall detection. This paper presents a cost-effective, robust solution for a fall detection system with the post-fall health tracking facility using a 3.18 GHz continuous-wave Doppler radar sensor. The experimental data acquisition is conducted in-house under the guidance of a healthcare expert, involving various activities such as standing, sitting, sleeping, running, walking, falling, sit-to-stand, and stand-to-sit transitions. We propose an algorithm comprising four hierarchical stages, each with specific objectives. Considering the complexity, the model is trained differently for each stage to optimize the classification accuracy. The system architecture is designed to minimize computational costs and power consumption through modular implementation in stages, utilizing low-power equipment and incorporating traditional machine-learning algorithms. Experimental results demonstrate a fall detection accuracy of 93.24% and breath rate measurement error of 2.26%, which is competitive with recent state-of-the-art approaches. Obtained results highlight the effectiveness of the proposed system in addressing the challenges of false alarms and post-fall health tracking while maintaining cost-efficiency and accuracy in fall detection.
现有的低成本基于多普勒雷达的跌倒检测系统由于误报和缺乏跌倒后健康跟踪而面临挑战，这极大地影响了它们的准确性和整体适用性。本文提出了一种使用 3.18GHz 连续波多普勒雷达传感器的具有跌倒后健康跟踪功能的低成本、鲁棒的跌倒检测系统解决方案。在医疗保健专家的指导下，在内部进行了实验数据采集，涉及各种活动，如站立、坐下、睡觉、跑步、行走、跌倒、从坐到站和从站到站的转换。我们提出了一个由四个层次结构阶段组成的算法，每个阶段都有特定的目标。考虑到复杂性，我们为每个阶段都训练了不同的模型，以优化分类准确性。该系统架构通过分阶段的模块化实现来最小化计算成本和功耗，使用低功耗设备并结合传统的机器学习算法。实验结果表明，跌倒检测的准确率为 93.24%，呼吸率测量误差为 2.26%，这与最新的最先进方法具有竞争力。所获得的结果突出了所提出的系统在解决误报和跌倒后健康跟踪挑战的同时保持跌倒检测的成本效益和准确性方面的有效性。
