Geometric Mean Based Log-Normal Distribution for Background Subtraction
Detection of moving object in dynamic, poor visibility environment is one of the challenging issues in today’s research work. The technique of Background subtraction is normally used for separation of foreground from background. In this research, we have used lognormal distribution of individual pixel as the features for background modeling. In our proposed method, we have used the concept of Geometric mean for averaging the lognormal distribution of individual pixel in temporal direction. Finally, the foreground has been separated from background by Otsu's method. We had extended our work, by separating the foreground from background by using log normal Otsu's method which proved to give better result as compared to the previous one. Both the proposed method has been tested in challenging environments in which high speed foreground objects like autonomous vehicles and pedestrian are been detected.