Real-Time Motion Up and Down Activity Recognition Based on Smart Phone and Smart Watch Sensors

Human activity recognition has been active research in the last few years. It can be used effectively in different applications such as fitness, games, indoor localization, healthcare, and emergency notification. Lacking highly accurate motion up and down human activities recognition and using traditional motion sensors in such systems encourage us to design and implement human activities recognition systems based on motion and environment sensors. A Dataset consisting of 10 subjects has been used in this work. Only one machine learning classifier DT has been used to predict four motions up and down activities, including upstairs, downstairs, elevator up, and elevator down, based on accelerometer, gyroscope, GPS, and barometer sensors. Ration-based feature extraction based on statistical features has been used with PCA for feature selections. The proposed system shows consistent classification performance with up to 98% accuracy. Our work mainly contributes to using whether a sensor as a motion sensor or a traditional motion sensor to increase system performance.