Conceptual group activity recognition model for classroom environments

With the development of smartphones containing built-in sensors of various kinds, an increasing amount of research effort is being devoted to recognition using wearable devices. In this paper, we limit our research to personal activity recognition, which is important to efficiently accumulate sensor data. We propose 1) a method to recognize conceptual group activity, and 2) a big data model to analyze large amounts of streaming data. This study focuses on three activities in the classroom environment: Taking a Lesson, Presentation, and Discussion. In our experiments, the proposed recognition algorithm recorded an accuracy of over 96%. We used the big data programming model MapReduce to accumulate and analyze data, and stored the sensor data and the activity data in a big data repository. In future research, we plan to study group activity recognition in other environments, and design a big data streaming system for group activity recognition.