Multi-Resident Activity Recognition with Unseen Classes in Smart Homes
Wei Wang, Chunyan Miao · 2018
Multi-resident activity recognition is important to many applications in smart homes. Existing works in this area are usually based on supervised classification methods, and can only recognize the predefined activity classes that are labeled in the training data. However, in many practical applications, the activities needed to recognize contain not only the predefined activities but also the previously unseen activities. In this paper, we propose a method to solve the problem of multi-resident activity recognition with the previously unseen activity classes. Our method utilizes the techniques of multi-task learning and zero-shot learning. It regards the activity recognition of each resident as a learning task, and learns all tasks jointly. By utilizing zero-shot learning techniques, previously unseen activity classes can be recognized. We conduct extensive experiments on real-world datasets. The experimental results show the effectiveness of our method.