Filtro de partículas hibridizado com métodos da computação natural para detecção e rastreamento

Abstract

Detecting and tracking objects in image sequences currently appears in various situationsof everyday life and stands out for its importance in many areas, for example, in security (monitoringobjects or persons), among others. A commonly used method is the Particle Filter,the main issue of Particle Filter is degeneration, which may imply a worse tracking. In this work, it is presented two hybrid method of Particle Filter. This hybridization occurs combining a Particle Filter and a natural computing: i) Particle Swarm Optimization; and ii) Differential Evolution. That way, aiming to minimize the degeneration problem in Particle Filter, in order to improve the performance of the tracking method. The proposed methods were applied to two case studies: i) for tracking the trajectory of the truck-trailer system, and ii) to detect and track a person s face in an image sequence. The results in terms of tracking quality indicate a better performance of hybridized algorithms when compared with the standard Particle Filter.