View reconstruction from images by removing vehicles

Reconstructing views of real-world from satellite images, surveillance videos, or street view images is now a very popular problem, due to the broad usage of image data in Geographic Information Systems and Intelligent Transportation Systems. In this paper, we propose an approach that tries to replace the differences among images that are likely to be vehicles by the counterparts that are likely to be background. This method integrates the techniques for lane detection, vehicle detection, image subtraction and weighted voting, to regenerate the "vehicle-clean" images. The proposed approach can efficiently reveal the geographic background and preserve the privacy of vehicle owners. Experiments on surveillance images from TrafficLand.com and satellite view images have been conducted to demonstrate the effectiveness of the approach.