Video Feature Representation and 3D Curve- based Event Matching

This paper highlights the progress of the research programme for investigating spatio-temporal volume based video event detection. The main research aim is to devise innovative and efficient volume-based video content analysis techniques and systems. To tackle this challenge, this paper fouces on the subjects of volume based feature analysis and event template matching. Corresponding system prototype design and experimental result analysis have also been presented in this report. The experiment result has clearly demonstrated the advantages of the volume based event detection methodology in terms of its richness of temporalrelated information and the potentials for identifying complex activities. As an important application, the difficulties in representing/extracting dynamic features from randomly ordered stacks of 2D snapshots can now be readily denoted as voxels for analysis in a 3D volumetric space. Keywords-event detection; volume processing; video processing; feature extraction