Activity and identitiy recognition from wearable sensors

Remotely detecting an activity and the person who performs this activity is an important issue that is needed in various fields. For this purpose, the usage of wearable motion sensors has been widespread in recent years. In this thesis, motion and person recognition were studied by means of accelerometer, gyroscope and magnetometer. The time, frequency and wavelet features were extracted from the data obtained from the sensors and learning algorithms such as Random Forest, J48, Adaboost and Desicion Stump, Support Vector Machine and k-NN were used for classification purposes. In addition, in order to improve the classification performance obtained; filtering, feature selection, fusion of sensors have been tried. The methods mentioned have been tried on both the open access data sets and hand activity data collected within the scope of this study, and the results have been reported.