A Comparison of Line Detectors for Image Background Modelling

In the context of multimedia data analysis, it is often necessary to understand what happens in a scene, which can rely on detecting and tracking the different moving objects and by comparing their positions at a given time with their environment. In order to model the environment or the scene background, we consider straight lines as appropriate features. In this paper, we compare some classical line detection approaches with two low cost contributed techniques, particularly relevant for real time scene analysis. The comparison methodology involves the definition of some evalutation measures. Results obtained show that a contributed method, called the connection approach, gives the best results.