Trajectory representation using Gabor features for motion-based video retrieval

摘要: This paper proposes a Gabor filter based representation of motion trajectory, for the purpose motion-based video retrieval. We propose spectro-temporal which involves process detecting set salient points from peaks (locally) responses. The change in trajectory direction is also represented by observing clockwise or anti-clockwise at points. feature (formed frequency, temporal location and turning each point) provides semantic trajectory. Our approach global where matching performed on edit distance shown to perform well even partial matching. system tested using two benchmark databases trajectories, as various hand-drawn trajectories. have experimented real world videos. Experimental results better performance than existing systems Fourier descriptors, polynomial state-of-the-art methods symbolic representations PCA characteristics movement.