Exploring appropriate clusters in subspace for human activity recognition

Activity recognition, which has emerged as a pivotal research topic in pervasive sensing over the last several years, utilizes a collection of data from sensors to capture human behavior, detect anomalies and provide warning or guidance information. This paper presents an approach to explore appropriate clusters in subspace for human activity recognition. The approach includes two major phases: discovery of human activity (extraction of human behavior patterns and generation of human activity clusters), and recognition of human activity (application of similarity function to recognize activities). Different from many existing works, the proposed approach applies a subspace clustering based algorithm to generate clusters of human activity. This approach aims to accumulate human activity by approximating the generated clusters to the activity from a conceptual human perspective. The experiments were implemented using radio-frequency identification (RFID) based systems. The results show that the proposed approach is effective in improving the accuracy of both activity discovery and activity recognition.