Presents a method for matching sets of trajectories whichsupplements motion information with knowledge about the spatialrelationships between points on the moving object. First the authorspresent a simple algorithm which matches two single trajectories usingonly motion information. They convert the 2D motion trajectories intotwo 1D signals based on the speed and direction components. The signalsare then represented by scale-space images both to simplify matching andbecause the scale- space representations are translation and rotationinvariant. They extend the matching algorithm to include spatialinformation and propose a second algorithm which matches multipletrajectories by combining motion and spatial match scores. Bothalgorithms were tested with real and synthetic data
展开▼