Design and Implementation of a Distributed Fall Detection System—Personal Server

Abstract

In this paper, the main results related to a fall detection system are shown by means of a personal server for the control and processing of the data acquired from multiple intelligent biomedical sensors. This server is designed in the context of a telehealthcare system for the elderly, to whom falls represent a high-risk cause of serious injuries, and its architecture can be extended to patients suffering from chronic diseases. The main design issues and developments in terms of the server hardware and software are presented with the aim of providing a real-time analysis of the processed biosignals. As a result, the evaluation study of the implemented algorithm for fall detection through a set of laboratory experiments is presented, together with some important issues in terms of the device's consumption. The proposed algorithm exhibits excellent outcomes in fall detection.

Ambient Assisted Living: A methodological approach

“…A deeper analysis of system architecture falls out of the scope of this paper. Taking into account the service deployment of the proposed architecture, although [26] shows a preindustrial prototype implemented by the authors, the technology independence of the PSE software design permits to be easily installed in several development hardware platforms, in agreement with the already referred new trends in embedded operating systems for wearable devices that ease SOA service deployment. This technology independence gives openness to the proposed architecture.…”

An unsupervised acoustic fall detection system using source separation for sound interference suppression

“…We need to remark that with the aid of the OCSVM algorithm, only the supporting vector (the sample x i whose corresponding coefficient α n i 4 0) is kept for constructing the decision function in (6). In this way, the decision function is not represented by all the training samples as in the traditional kernel based method [3] and the computational cost for deciding whether a data sample is normal/abnormal could then be reduced.…”

“…The results showed that fall detection using a triaxial accelerometer worn at the waist or head together with a simple thresholdbased algorithm is efficient, with a sensitivity of 97-98% and a specificity of 100%. Acceleration sensors can be used with other devices to achieve a comprehensive fall detection system, in Estudillo-Valderrama [6], a low-power waterproof biocompatible accelerometer smart sensor (ACSS) was applied and an additional user interface module was integrated in the second layer (denoted as personal server (PSE) in this paper) to allow the elderly person to access some of the most important data being processed; from the algorithm aspect, an additional time analysis was used by convolving the resulting acceleration data segment with certain defined waveforms, to detect some problematic fall events such as a knee fall. A total of 332 samples of fall and non-fall activities simulated by 31 young and healthy males and females were tested, 100% sensitivity and 95.68% specificity were obtained and a further reduction of false positives can be obtained by manually canceling the fall alarm through the user interface.…”

Development of Wearable Human Fall Detection System Using Multilayer Perceptron Neural Network

“…The window size for preprocessing is very important in detection accuracy because a small window will misclassify patterns while a large window has the potential to misclassify a fall and has low performance in fast movement recognition. 16 Therefore, two different experiments were performed in order to find the best window size. In the first experiment which is a statistical analysis of the recorded acceleration database, the average falling duration time was calculated.…”