Automatic path modeling by image processing techniques

In recent years, many studies have focused on intelligent video surveillance system, including camera calibration, foreground region detection, moving object detection, moving object tracking and path modeling. This study used the data of the moving trajectory of a moving object as the path modeling data. However, the data may contain incorrect trajectory data, such as wrong foreground region detection data, wrong moving object tracking data, or moving object trajectory not on the normal path, thus resulting in incorrect path. This study first used Background Subtraction to capture moving objects, such as pedestrians or vehicles from the video, and then applied Morphology Operation and Connected Components to eliminate noise and label every individual moving object. Finally, gravity center of each moving object was calculated to obtain the path modeling data. Different from previous path modeling, this study used reward and punishment mechanism to automatically adjust path modeling weight, thereby reducing the impact of inferior trajectory on path, and improving the path model performance with the new path.