Abstract: Current image-based road crack recognition can only obtain local crack segmentation detection results, it is difficult to obtain the actual location information of this crack, and it is difficult to automatically quantify the actual size of all cracks in the overall structure. To solve these problems, a road crack localisation and quantification method is proposed based on UAV monocular video. In the framework proposed, the image data used for road reconstruction is firstly optimized using the key frame selection principle in ORB-SLAM3 to achieve the same quality reconstruction with less number of images; then an improved SegFormer semantic segmentation model is used for prediction, and the semantic information such as cracks and construction joints are spliced together to form a crack map of the whole road; finally, the Alpha-shape algorithm is used to complement the road profile, and crack skeleton extraction, according to the specification to automatically calculate the pavement surface condition index of each unit. Tests are carried out on a road in Changsha to verify the effectiveness of the method proposed, and the study results show that, this method can obtain acceptable accuracy with a small amount of image annotation, complete the whole road crack identification and localization with a high degree of automation, and significantly improve the efficiency of road crack detection.