An image processing device includes an acquisition unit, an extraction unit, and a detection unit. The acquisition unit acquires image data for each of frame numbers from a drive recorder or the like. The extraction unit extracts a plurality of straight lines from the image data on the basis of drive recorder information. The detection unit detects a tilt of an image capturing direction of a camera according to a magnitude of an angle between a vertical axis direction of the image data and the extracted plurality of lines.