Smart fall detection monitoring system using wearable sensor and Raspberry Pi

• Published in: AIP Conference Proceedings
• Main Author: Lim C.C.; Mahmud N.F.A.; Vijean V.; Ali Y.M.; Salleh A.F.; Tan X.J.; Basah S.N.
• Format: Conference paper
• Language: English
• Published: American Institute of Physics Inc. 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85185768299&doi=10.1063%2f5.0192471&partnerID=40&md5=de69d4056ac639f71610a3e500ac58ea

The Smart Fall Detection Monitoring System is the name of the programme that monitors everyday activities and falls. It has an accelerometer sensor (ADXL345) and Raspberry Pi 3 microcontroller board to recognise and classify the patient's fall. Python programming was done on the Raspberry Pi terminal to enable communication between the accelerometer sensor and the computer. There were 10 subjects (5 males and 5 females) collected. While daily living activities include standing, squatting, walking, sitting, and lying, the data on falling includes forward falls and falls from medical beds. The K-nearest Neighbour (kNN) classifier can categorise the data of falling and non-falling (everyday living activity). The accuracy of the kNN classifier was 100% for the combined feature and (>87%) for each feature during the categorization of the falling and non-falling classes. In the meantime, multiclass classification performance for combining features and for each feature separately was >85%. kNN classifier was used to assess the feature. The feature was chosen based on the k-NN classifier's accuracy score as a percentage. For feature selection for falling and non-falling, feature (AcclX, AcclY, AngX, AngY and AngZ) in City-block distance was selected as they performed high accuracy which was 100%. The performance of the AngZ (77%) was good during the sub-classification of the sub-class dataset. As a result, all feature characteristics were chosen to be incorporated in the IoT fall detection device. The system is real-time communication for classifying fall and non-fall conditions with 100% accuracy using kNN classifier with cityblock distance. © 2024 Author(s).