A road map generation system is provided that collects camera image data of road conditions captured during traveling of vehicles on roads and generates road map data based on the camera image data. The system performs image processing on the collected camera image data to convert into orthographic images and to combine the orthographic images to generate a combined image. The system extracts a defective portion of a road marking on a road from the combined image and performs a determination of whether the defective portion of the road marking is due to noise at the image processing of the image processing device or there is a defect in the real world. When the defective portion of the road marking is due to the noise at the image processing, the system performs a correction process of the road marking.