Adaptive Foreground Object Detection in Railway Scene

In modern railway systems, intelligent video analysis has been widely applied in foreign body intrusion monitoring, and foreground object detection is an essential step for intrusion detection. Background subtraction is commonly used to detect foreground objects. However, existing threshold-segmentation-based methods and deep-learning-based methods cannot meet the requirements in a complex railway scene, which contains dynamic background and unknown objects. In this paper, we proposed an adaptive-threshold-based foreground detection algorithm, which utilizes the temporal dynamic of pixel intensity, feedback information of detection result and spatial information of super-pixel to determine a factor, and then automatically adjusts the threshold by the factor to follow scene change. In addition, we also proposed a flexible and reliable background model initialization method that eliminates the ghost problem and flexibly switches from one-frame initialization to multiple-frame initialization. Experimental results show that the proposed algorithm achieves better accuracy and wrong classification rate in railway scenes, and also gets a better trade-off between accuracy and speed. Copyright © 2020 by Science Press.