Multisenzorsko izdvajanje i ažuriranje slike pozadine u detekciji pokretnih objekata / Multisensor background extraction and updating for moving target detection

Abstract

U radu je predložen postupak multisenzorskog izdvajanja i ažuriranja slike pozadine koji se koristi za video nadzor i detekciju pokretnih objekata. Izdvajanje slike pozadine zasnovano je na detekciji promena na slici i iterativnom postupku ažuriranja. Pri detekciji promena koriste se informacije raspoloživih senzora, čime se povećava pouzdanost detekcije i slike pozadine. Predloženi postupak testiran je na bazama registrovanih televizijskih i termovizijskih slika. / This paper presents a robust multisensor background extraction and update scheme intended for use in target detection applications. The method uses a number of well known concepts in detection to provide a basic classification of the scene into moving and stationary objects and an iterative procedure to update the background model and deal with transient (slow) changes in illumination conditions. A new fusion- based confidence framework uses the information from multiple registered views of the scene obtained by different sensors in order to update the background model only in areas where there are no moving objects. It was shown on a number of representative multisensor sequences that the proposed method obtains a more complete and robust representation of the background than any estimate based on the individual sensors. Finally, a relatively simple computational architecture of the proposed system makes it ideal for demanding real time multisensor detection applications.

Keywords

• multisenzorska detekcija
• izdvajanje slike pozadine
• inicijalizacija i ažuriranje slike pozadine
• multisensor detection
• background extraction
• background initialization and updating