Posture Recognition with G-Sensors on Smart Phones

With the popularity of smart phones in recent years, various sensors on smart phones can be utilized to detect the movement or intention of the smart phone users. In this research, we aim at using the signals collected from the G-sensor in the smart phone to recognize the posture of the user. Signals for sit, stand, walk and run are collected to train an offline neural network as the classifier. After the neural network learns the four postures, we then implement a neural network with the learned connection weights in a smart phone app. The app can record the postures of the user for the whole day and estimate the burned calories accordingly. This app can replace the pedometer to have a more accurate estimate of calorie consumption. Details of the app are presented in this paper. The accuracy of neural networks on posture recognition with G-sensor signals is also verified by five-fold cross-validation.