Comparison of Methods for Determining Activity from Physical Movements

In this study, the methods which can detect basic physical movements of a person (downward, upward, sitting, stop, walking, running) from inertial sensor data are evaluated. The performances classical and current approaches to map IMU activity classes were compared. A three-stage study was carried out for aim: 1) acquisition; 2) creating training/test sets; 3) construction classification network architectures. At stage acquisition, obtain 6 different 10 people, accelerometer is placed on persons. Repetitive persons recorded. second stage, recorded long-term divided into packages in form short-term windows. training set constructed by features extracting each packet containing one-dimensional acceleration information. transformation signals two-dimensional image matrix deep learning-based performed. third ANN, SVM, k-NN CNN architectures constructed, process out. As result experimental studies, it found that accuracy IMU-activity mapping 99% with ANN method 95% method.