Camera-less context detection in mobile platforms

Ever since their introduction, smartphones have become more and more powerful, and offer an ever increasing array of sensors for developers to use. This however comes at the cost of the device’s battery life. The goal of this study is to evaluate the feasibility of implementing a deep learning classifier on an Android device to save power when the device changes from an “in use” state to a “not in use state”. To this end, this work explains how data was collected to train a Convolutional Neural Network for the task, how the neural network was trained and how parameters were selected, and an analysis on which sensors were most important for the classification, and which could be removed in a later work. Results show that a Convolutional Neural Network works very well for this case, and proves that with just 2 seconds of data collected from few low-powered sensors, in 96% of the cases, the classifier will recognise that the device is not in use. This would allow for large amounts of energy savings compared to a simple timeout, while still consuming a negligible extra amount of power for the classification itself.