Matching motion trajectories using scale-space

摘要: Abstract The goal is to design a recognition system which can distinguish between two objects with the same shape but different motion, or motion shape. input set of two-dimensional (2D) trajectories from an object tracked through sequence n frames. structure and three-dimensional (3D) each in domain are stored model. problem match information model 2D determine if they represent object. simplest way perform these steps projections 3D trajectories. First, simple algorithm presented matches single using only information. converted into one-dimensional (1D) signals based on their speed direction components. then represented by scale-space images, both simplify matching because representations translation rotation invariant. extended include spatial second proposed multiple combining scores. Both algorithms tested real synthetic data.