Motion Sequence-Based Human Abnormality Detection Scheme for Smart Spaces

Abstract

Smart spaces represent an emerging new paradigm that encompasses diverse active research areas such as ubiquitous, grid and cloud computing. Hence, there are a wide variety of interesting issues and applications for smart spaces, and surveillance is one issue that has long received much attention. In many cases, human motion is one of the most important clues used in assessing a situation for surveillance purposes. In this paper, we propose a new human abnormality detection scheme for surveillance purposes. More specifically, we first present a motion sequence matching algorithm called Dynamic View Warping to represent specific motion characteristics. Secondly, we propose a matching speed-up technique called Dynamic Group Warping that establishes boundaries in Dynamic View Warping. Thirdly, we propose an indexing scheme for motion sequences and present K-NN search algorithm to efficiently and effectively find similar motion sequences. Our extensive experiments show that our proposed methods achieve outstanding performance.