Moving object detection in airborne video using kernel density estimation

A method to detect moving object in airborne video based on kernel density estimation was proposed. First, the global motion of background was estimated according to the corresponding control points. In order to increase the accuracy of the motion estimation, the relative position of the feature points was used to remove the invalid point pairs. After background compensation, the background model was constructed by use of nonparametric kernel density estimation with a sample of image sequences where the background was stationary. The moving objects were obtained through background subtraction. The spatial errors of background compensation were analyzed to eliminate the false detection mainly caused by background compensation. Finally, the detection result of background subtraction and successive inter-frame motion information were used for background updating. Therefore, the background model can be adaptive to the change of object motion. Experimental results show that the proposed method is effective for moving object detection in airborne video.