Please use this identifier to cite or link to this item:
|Type:||Artigo de evento|
|Title:||Distributed Real-time Soccer Tracking|
|Author:||Muller Jr. B.|
De Anido R.O.
|Abstract:||Tracking objects that take part in sportive events is a challenging task because the objects move fast and occlusions occur frequently. When the tracked area is large, the use of more than one high resolution cameras improve accuracy, but leads to a huge amount of data to be processed and fused. The cameras are usually placed to maximize the covering area, and thus the tracked objects are small, usually 10 to 40 pixels height. This paper presents a new approach to this kind of application, where the tracking procedures are not applied to the whole images, but to small images taken from the cameras. Given a specific location of the tracked area, the system is able to return a set of small images (say 60×60 pixels) centered on that location, one from each camera, and the tracking procedures are applied to these images. Each object can be tracked individually by an independent module of the system, and each module can apply different tracking techniques depending on specific visual characteristics. The paper describes a real-time distributed implementation of such system, and presents a new mechanism to detect objects in small images using a gradient reference frame. Copyright 2004 ACM.|
|Appears in Collections:||Unicamp - Artigos e Outros Documentos|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.