| dc.description.abstract | This work aims to present a model for detection of unusual motion based on trajectories. This model relates to the research field on intelligent cameras and surveillance systems, that tends to compete nowadays with the enormous range of devices based in hardware available on the market. The main idea of the proposed approach is to analyze trajectories acquired from film scenes. The first step of the algorithm consists of a training period, that learns the profile of trajectories, selecting, grouping and later, keeping them in a database. After that, the algorithm compares new trajectories that are being acquired continously in the test period. In test period, one given trajectory will be classified as usual if it is compatible with the trajectories acquired during the training period, or unusual otherwise. This work, therefore, will present algorithms that detect patterns of similarity between a set of trajectories in the training period with each new trajectory acquired in the test period | |
