A vision-based self-localization method was proposed for underground mining vehicle by using landmarks. Firstly, several artificial landmarks were located in a simulating tunnel, and then a vision sensor in vehicle was used to find and recognize them. After the distances between the vehicle and the recognized landmarks were calculated by a visual distance measurement method, the location of the vehicle was computed according to triangulation finally. In this self-localization method, landmarks were encoded by interleaved 2 of 5, a type of 1-D barcode with high density. Each landmark has obvious visual feature and there is a unique corresponding location data to it in the landmark database. Distance between a landmark and the vehicle was calculated on the basis of the ratio of the coding area’s actual height to its height in the image according to the pinhole imaging principle. Experimental results showed that the self-localization method suggested almost meets the requirement for autonomous driving vehicle because of its high efficiency and precision.