Convolutional neural network-based tumble detection method

Abstract

The invention provides a convolutional neural network-based tumble detection method, and belongs to the field of electronic information. The method comprises the following steps: converting three-axisacceleration and angular velocity data acquired by an MEMS into RGB pixels, and introducing a sliding window to construct a pixel graph capable of simultaneously reflecting the change characteristicsof the three-axis acceleration and the angular velocity in the activity process of old people; designing a fall monitoring algorithm FD-CNN based on CNN by referring to LeNet 5 architecture, and classifying pixel images to realize a fall detection algorithm. By constructing the FD-CNN, interferences such as signal drift errors of the MEMS gyroscope and voltage fluctuation generated by the three-axis accelerometer in a motion state can be overcome, and fall detection is accurately realized. The accuracy of the FD-CNN network model system reaches 98.62%, and the sensitivity and the specificityof the FD-CNN network model system reach 98.65% and 99.80% respectively. The sensitivity and specificity of the system to fall detection both reach 100%.

Description

technical field [0001] The invention belongs to the field of electronic information, and is a technology based on a convolutional neural network that can be applied to fall detection. Background technique [0002] The existing fall detection technologies can be generally divided into the following three categories according to the different sensor technologies used: Wearable Sensors, Ambient Sensors and Vision-based Sensors. Although situational awareness technology and visual perception technology have the characteristics of high accuracy and intuitive motion capture, sensor deployment and detection algorithms are complex, and the monitoring range is limited, which may even easily expose user privacy. With the development of MEMS (microelectromechanical systems) technology, researchers have integrated inertial sensors into small wearable devices for fall detection. The fall detection method based on MEMS has the advantages of low deployment cost, less environmental impact,...

AI Technical Summary

Problems solved by technology

However, there are signal drift errors in the MEMS gyroscope, and voltage fluctuations in the 3-axis accelerometer during motion. These factors will affect the accuracy and effectiveness of the fall detection algorithm

Images