Video volume segmentation for event detection

Video processing for surveillance and security applications has become a research hotspot in the last decade. This paper reports a research into volume-based segmentation techniques for video event detection. It starts with an introduction of the structure in 3D video volumes denoted by spatio-temporal features extracted from video footages. The focus of the work is on devising an effective and efficient 3D segmentation technique suitable to the volumetric nature of video events through deploying innovative 3D clustering methods. It is supported by the design and experiment on the 3D data compression techniques for accelerating the preprocessing of the original video data. An evaluation on the performance of the developed methods is presented at the end.