Intelligent video object tracking in large public environments

• Autor(a) principal: Ferreira, Marco Paulo Fernandes
• Data de Publicação: 2009
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/23212
• topic: Multiple video object tracking; Occlusions handling; People tracking; Real-time video surveillance; Seguimento de múltiplos objectos de vídeo; Processamento de oclusões; Seguimento de pessoas; Videovigilância em tempo real

This Dissertation addresses the problem of video object tracking in large public environments, and was developed within the context of a partnership between ISCTE-IUL and THALES1 object. This partnership aimed at developing a new approach to video tracking, based on a simple tracking algorithm aided by object position estimations to deal with the harder cases of video object tracking. This proposed approach has been applied successfully in the TRAPLE2 project developed at THALES where the main focus is the real-time monitoring of public spaces and the tracking of moving objects (i.e., persons). The proposed low-processing tracking solution woks as follows: after the detection step, the various objects in the visual scene are tracked through their centres of mass (centroids) that, typically, exhibit little variations along close apart video frames. After this step, some heuristics are applied to the results to maintain coherent the identification of the video objects and estimate their positions in cases of uncertainties, e.g., occlusions, which is one of the major novelties proposed in this Dissertation. The proposed approach was tested with relevant test video sequences representing real video monitoring scenes and the obtained results showed that this approach is able to track multiple persons in real-time with reasonable computational power.