Motion analysis using 3D Gabor kernels

Today human motion understanding has become one of the most active research areas in computer vision. This is due to promising applications in areas such as visual surveillance, human performance analysis, computer-human interfaces (robotic interaction with humans), content-based video retrieval/storage and virtual reality. This work is focused on recognizing human actions. A novel motion analysis method is proposed. Gabor kernels, which are appears to be the most successful approach for 2D texture segmentation and modeling, are extended to the time domain to extract temporal texture features as well. Human motion is represented by states and a dynamic time warping based approach is used for motion matching.