Adaptive window based fall detection using anomaly identification in fog computing scenario

Abstract

Human fall detection is a subcategory of ambient assisted living. Falls are dangerous for old aged people especially those who are unaccompanied. Detection of falls as early as possible along with high accuracy is indispensable to save the person otherwise it may lead to physical disability even death also. The proposed fall detection system is implemented in the edge computing scenario. An adaptive window-based approach is proposed here for feature extraction because window size affects the performance of the classifier. For training and testing purposes two public datasets and our collected dataset have been used. Anomaly identification based on a support vector machine with an enhanced chi-square kernel is used here for the classification of Activities of Daily Living (ADL) and fall activities. Using the proposed approach 100% sensitivity and 98.08% specificity have been achieved which are better when compared with three recent research based on unsupervised learning. One of the important aspects of this study is that it is also validated on actual real fall data and got 100% accuracy. This complete fall detection model is implemented in the fog computing scenario. The proposed approach of adaptive window based feature extraction is better than static window based approaches and three recent fall detection methods.

Deep Transfer Learning Driven Automated Fall Detection for Quality of燣iving of Disabled Persons

“…Further, it also reduces certain requirements for human accommodations or intermediaries. Nowadays, advanced IoT devices and their services can be easily accessed by disabled persons; some IoT techniques are specially developed for disabled people, whereas a few other techniques are re-processed for disabled individuals [2]. The IoT and the associated information collection result in availabilitybased developments ranging from smart home devices to self-driving cars [3].…”

Edge device for movement pattern classification using neural network algorithms

“…The main contributions of the study are: i) an experimental work is shown to analyze the behavior of different machine learning models in an edge device with and without the use of a quantization framework, in the context of human activity recognition [19]- [21] using an own dataset; ii) the edge device is evaluated by analyzing inference time, memory usage and classification accuracy for two types of neural network-based learning models. In addition, the results of the paper allow to obtain indications to contribute to future research with machine learning solutions integrated in edge devices.…”