Living activity recognition using off-the-shelf sensors on mobile phones

摘要

In aging societies, such as that of Japan, there is growing awareness that robotic technology has the potential to help both physical and mental labor. To take an example of mental labor, the robotic technology can contribute as an interface to home electric appliances and a conversation partner with interactive communication. In this case, it is important to recognize the elderly user's activities for not only watching-over services but also improving the quality of the conversation. We propose a low-throughput recognition method for in-home living activity recognition using only off-the-shelf sensors, namely an accelerometer and a microphone, which are commonly applied in mobile phones. The system can determine whether the user is walking, quiet, or performing a task by acceleration sensing, and then in the latter case, acoustic sensing can be used to classify the nature of the task that the user is performing. We conducted two experiments to confirm the feasibility of the proposed method. As a result of the first experiment, three movement conditions are classified with more than 95 % accuracy by acceleration sensing: walking, quiet, or performing a task. And it classified the nature of the task into brushing teeth, shaving, drying the hair with a hairdryer, flushing the toilet, vacuuming, washing the dishes, and ironing with 75.8 % accuracy by acoustic sensing and improved the accuracy to 85.9 % by training with only the subject's own data. Moreover, the result of the second experiment shows that it is effective to adopt instance- based recognition which is an additional recognition scheme per each continuous task, according to the assumed application.