A template matching is performed on two fluoroscopic images by using a template image prepared in advance and a position corresponding to a high matching score is listed as a candidate for the position of a marker 29. From two lists of the candidates of the position of the marker 29, the lengths of common vertical lines for all combinations are calculated. Then, the position of the marker 29 is detected based on the matching score and the common vertical line. Then, based on the detected position of the marker 29, an amount of a proton beam to be irradiated to a target is controlled. Therefore, a tracking target can be accurately detected even when the conditions for X-ray fluoroscopy is severe, e.g., a thick object.