Neural network based daily activity recognition without feature extraction

In recent years, human-computer interaction systems are become one of the most exciting areas in technological development. These systems aim to obtain personal information of people and development of an automated systems managed by this information. In this study, we have been studied a faster and higher accurate system design without feature extraction for the recognition of daily human activities and falling situation. Motion data were collected under knee with a 3-axis accelerometer. After data re-arrangement, a 250 data size window was applied to collected data. 250 XYZ axis data belonged to each windowed sample were written in an array and converted to 1×750 sized array. Finally, applying data reduction with PCA, the data were simulated by MLP, SVM and Naive-Bayes classifiers. The best result without feature extraction achieved by Naive Bayes classifier.