Fall Detection and Prediction Based on IMU and EMG Sensors for Elders

Abstract

One of the most crucial changes in the future of social structure is the increase of the aging population. Accidents in the elderly are often caused by degeneration and worsening of their bodies. The most common accidents in the elderly are falls. This research proposes fall prediction and detection methods based on the Inertial Measurement Unit (IMU) sensor and Electromyogram (EMG). This method used features from EMG signal to adjust and co-verify with IMU sensor and then used machine learning technicians to create the model for abnormal classification gait, normal gait, and fall event. The results show that the EMG signal based on the Random forest model gained the average accuracy values of 3-class classifications (Abnormal gait, Normal gait, and Fall event) is 71.91%. For 4-class classifications (Abnormal left leg, Abnormal right leg, Normal gait, and Fall event) is 67.76%. The IMU sensor base on the Random Forest (RF) model got the best performance on both accuracies at 3-class and 4-class classification; the average accuracy value of 3-class classification is 94.72%. For the 4-class classification is 87.70%, respectively.