Lane line detection system based on geometric attention perception

Abstract

The invention provides a lane line detection system based on geometric attention perception. The lane line detection system a backbone network;a semantic segmentation branch and ageometric distance embedding branch which are arranged behind the backbone network; an attention information spreading module which acts between the sampling layers in two adjacent stages of the decoder, namely is arranged between the whole lane line semantic segmentation branch and the geometric distance embedding branch; a geometric attention sensing module which is arranged at the tail ends of the semantic segmentation branch and the geometric distance embedding branch; and a jump pyramid fusion up-sampling module which is connected with the backbone network and the geometric attention sensing module. The system adopts a multi-task branch network structure; in addition to a lane line segmentation task, a geometric distance embedding branch is added, and the branch guides lane line segmentation by learning continuous distance representation from the center of a lane line to a boundary, so that the problem that the lane line cannot be effectively detected from a complex road scene due to the fact that theheight depends on sparse lane line labeling is solved.

Description

technical field [0001] The invention relates to the technical field of lane line detection, in particular to a lane line detection system based on geometric attention perception. Background technique [0002] Lane line detection is a technology that extracts static lane line features from images of the car's surroundings captured by on-board sensors. There are only a few methods to detect lane lines from images captured by lidar sensors, although lidar has natural advantages in distance measurement , but its imaging principle determines that it can only perceive lane lines with obvious signs, and the cost of lidar is very high. Therefore, low-cost cameras are used in most work. Because its imaging principle is similar to that of the human visual system, the images it captures are more suitable for human understanding, and it is convenient for manual labeling of images, including some complex traffic environments. For example, lane lines are Occlusion, etc., which are crucia...

AI Technical Summary

Problems solved by technology

[0009] Aiming at the technical problems existing in the prior art, the present invention provides a lane line detection system based on geometric attention perception. The system adopts a multi-task branch network structure. In addition to the lane line segmentation task, a geometric distance embedding branch is added. The branch By learning the continuous distance representation from the center of the lane line to the boundary to guide the lane line segmentation, it can improve the problem that the lane line cannot be effectively detected from the complex road scene due to the high dependence on the sparse lane line labeling.

Images