Um ambiente distribuido para rastreamentos de objetos em tempo real

AUTOR(ES)
DATA DE PUBLICAÇÃO

2004

RESUMO

This thesis deals with the problem of tracking objects from multiple static cameras. Each camera covers a specific area where the events take place, and the union of the images of the cameras cover the whole tracking area. The thesis presents a system that divides the tracking in two modules. The first module detects all the objects present in the tracking area from time to time, and to do so, it uses one frame taken from each camera at the same instant in time. The second module applies the tracking procedures to all frames, but work with smaller images taken from each frame. These smaller images are centered on the tracked objects. This approach has two main advantages, it reduces the amount of data to be analysed, is well suited to be implemented in a distributed environment. 1t also presents some new challenges, since the conventional approach is to apply the tracking procedures to the full frames. For this reason, some new techniques to track objects are introduced. To detect the objects in small frames, two algorithms are described. The first is based on the color of the tracked object and enhances the histogram back projection algorithm to be used in video sequences. The second algorithm uses a gradient reference frame, where there is no need for frequent updates. In addition, algorithms to remove shadows, to deal with occ1usion, to fuse objects that are visible in more than one camera and to track along time are also presented

ASSUNTO(S)

reconhecimento de padrões processamento de imagens rastreamento automatico

ACESSO AO ARTIGO

Documentos Relacionados