Killer heuristic optimized convolution neural network-based fall detection with wearable IoT sensor devices

Abstract

Current multi-disciplinary research views falls in the elderly as a significant worldwide public health risk. Several fall detection systems have been created that use wearable motion sensors, but these systems fail to provide an accurate assessment of the exact nature of human falls. This paper introduces an effective and optimized fall detection system that uses an approach based on a killer heuristics optimized AlexNet convolution neural network. Wearable sensor devices, which consist of a magnetometer, gyroscope, and accelerometer tri-axial device, are placed on the subject’s body in 6 different positions. During the data collection process, 16 activities of daily living and information on 20 voluntary falls are collected in 2520 trials. Information is collected from the IoT-assisted wearable device for feature extraction and analysis of sensor data. The derived features are analyzed by multilinear principle component analysis, which reduces the dimensionality of the features. Fall detection is then performed by applying the intelligent AlexNet convolution network. The fall detection based on the created wearable sensor device is evaluated using simulation results, and the system recognizes a fall with maximum accuracy and minimum complexity.