Labeling method for road image ground marking

Abstract

The invention discloses a road image ground marking method, relates to the technical field of vehicle automatic driving, and can quickly and accurately mark pixel geometric information of a road marking in an image when a large amount of image data is marked. According to the semi-automatic marking method, the edge points on the two sides of the marking line can be automatically obtained only by clicking any point near the pixel of the marking line, the complete pixel geometric information of the whole marking line can be obtained after multiple times of continuous clicking, accurate manual pixel point-by-pixel selection is not needed, the semi-automatic marking method greatly reduces the marking difficulty, and the marking efficiency is improved. The marking speed and precision are greatly improved; besides, the method is higher in adaptability, not only can be used for marking transverse marking lines in the image, but also can be used for marking longitudinal marking lines in the image, not only can be used for marking lane lines, but also can be used for marking various ground marking lines such as stop lines, pedestrian crossing lines and guide lines, and not only can be used for marking white marking lines, but also can be used for marking yellow marking lines.

Description

technical field [0001] The invention relates to the technical field of vehicle automatic driving, in particular to a method for labeling road image ground markings. Background technique [0002] High-precision maps are an essential part of current driverless technology, and ground marking data such as lane lines are the core content of high-precision maps. Vehicles usually need to travel along the centerline of the lane determined by the two lane lines in the high-precision map. Automatic driving, so ground markings such as lane lines play a very important role in the functions of each module of unmanned driving. It is often necessary to identify the exact position of the lane line very accurately. In order to achieve these goals, a large amount of lane line image annotation data is required , used to provide learning for the algorithm model and improve algorithm capabilities. [0003] At present, the existing technology adopts a pixel-by-pixel manual marking method for gro...

AI Technical Summary

Problems solved by technology

[0003] At present, the existing technology adopts a pixel-by-pixel manual marking method for ground marking lines such as lane lines. The object of marking is to directly mark a pixel center line, or mark two side edge lines, or mark the outer contour of the entire marking line, one by one. The method of manually labeling pixels is very slow. In order to mark the accuracy, the image needs to be enlarged first, and then accurately manually select points pixel by pixel. Labeling an image is very time-consuming and laborious.

[0004] For marking lines with a fixed width such as lane lines, the existing scheme usually directly marks a center line to represent the entire marking line, but the centers of these white or yellow lines are judged by the naked eye, and the marking accuracy is difficult to guarantee. If you label the edge lines on both sides or the entire contour line pixel by pixel, the amount of labeling will increase greatly, and the labeling speed is very slow

Images