Bed-Attached Vibration Sensor System: A Machine Learning Approach for Fall Detection in Nursing Homes

TL;DR

This study addresses falls in nursing homes and privacy concerns by proposing a bed-attached vibration sensing system that uses a logistic regression pre-filter to trigger short-time Fourier transform analysis, feeding spectrograms into a convolutional neural network for fall classification. The approach combines a low-power hardware stack (Teensy 4.1 and AIS2IH MEMS) with a two-stage AI pipeline and data augmentation to compensate for limited lab data, achieving higher precision as augmentation increases while maintaining recall at 1.0. Field tests in nursing homes demonstrated real-world challenges and data gaps, underscoring the need for broader, more diverse data and real-time validation. Overall, the work shows promise for a privacy-preserving, rapid-response fall detector that can operate without wearables or video, laying groundwork for further AI-enhanced sensor development in ShapeFuture.

Abstract

The increasing shortage of nursing staff and the acute risk of falls in nursing homes pose significant challenges for the healthcare system. This study presents the development of an automated fall detection system integrated into care beds, aimed at enhancing patient safety without compromising privacy through wearables or video monitoring. Mechanical vibrations transmitted through the bed frame are processed using a short-time Fourier transform, enabling robust classification of distinct human fall patterns with a convolutional neural network. Challenges pertaining to the quantity and diversity of the data are addressed, proposing the generation of additional data with a specific emphasis on enhancing variation. While the model shows promising results in distinguishing fall events from noise using lab data, further testing in real-world environments is recommended for validation and improvement. Despite limited available data, the proposed system shows the potential for an accurate and rapid response to falls, mitigating health implications, and addressing the needs of an aging population. This case study was performed as part of the ZIM Project. Further research on sensors enhanced by artificial intelligence will be continued in the ShapeFuture Project.

Figures (10)

• Figure 1: Overview of the developed fall detection system. The upper section illustrates the hardware components, while the lower section details the machine learning approach. The implementation of an alarm system, user interface, and central inference platform remains outside the scope of this study.
• Figure 2: Structure of MEMS Sensors with Spring Element Based on the Resistive Principle (Middle) and the Capacitive Principle (Right).
• Figure 3: Positive amplification of a dummy event. The orange scatter plot shows the original data, and the blue plot represents the amplified signal. The x-axis indicates time steps, and the y-axis represents acceleration.
• Figure 4: Lab data acquisition setup. The Left and middle images show sensor placements, while the right figure displays the corresponding acceleration curves. The sensor on the slatted frame exhibits the smallest acceleration, indicating the lowest sensitivity.
• Figure 5: Structure of the CNN. The input has a shape of (63, 251) pixels, and 240 Conv2D filters of size (63, 145) are applied. The resulting output is reduced in dimension using a MaxPooling2D layer (1, 4), flattened, and then passed to the output layer.