A dynamic online background modeling framework for moving object detection from airborne videos

Current researches on moving object detection from airborne videos are mainly based on frame difference. Though many improvements have been made on these methods, it is still difficult to extract all the moving pixels accurately. Being capable of providing more reliable motion information, background subtraction based methods have been widely used for analyzing surveillance videos captured by fixed cameras. In this paper, we design a dynamic online background modeling framework to facilitate the adaption of the available background subtraction algorithms for moving object detection from airborne videos. It can avoid accumulated stabilization errors and handle the pixels near the frame boundary well. The advantage of our framework lies in the stabilization strategies we proposed and the background model size we employed. Experimental results and analysis on the airborne videos have validated the effectiveness of the proposed framework.