Deep learning-based lane line detection method and system

Abstract

The invention discloses a lane line detection method and system based on deep learning, and belongs to the technical field of automatic driving, and the method and system are high in lane line recognition, especially far-end lane line recognition accuracy, low in network bandwidth requirement and low in user privacy data requirement. The method comprises the following steps: comparing the real-time curvature radius of a lane line obtained by recognition of a terminal lane line detection model with the real-time curvature radius of the lane line obtained by a map engine, if an existing error is greater than a set value, correcting the curvature of the lane line obtained by a vehicle-mounted lane line detection model, and meanwhile, correcting the curvature of the lane line obtained by the vehicle-mounted lane line detection model; comprising the following steps: regularly transmitting the corrected actual position of the terminal lane line detection model, the corrected curvature and the decrypted user privacy image information to a background server, updating the parameters of the cloud lane line detection model, and then issuing the parameters to each lane line detection terminal; and if the error of the two curvature radiuses in the horizontal direction is smaller than a set value, the lane line obtained by the lane line detection model is regarded as a correct lane line.

Description

technical field [0001] The invention relates to a lane line detection method and system based on deep learning, and belongs to the technical field of automatic driving. Background technique [0002] Traditional lane line detection methods are usually based on visual information. Such methods use algorithms such as HSI color model and edge extraction to obtain visual information of the image. Tracking becomes another popular post-processing scheme when visual information is insufficient. Besides tracking, Markov models and conditional random fields are also used as post-processing schemes. With the development of machine learning, some schemes using model matching and support vector machines have also been proposed. [0003] With the development of deep learning, some deep neural network-based schemes have shown excellent performance in lane line detection. At present, the mainstream deep learning methods for lane line detection mainly include: semantic segmentation, sequ...

AI Technical Summary

Problems solved by technology

[0004] However, when the road conditions are complex, the accuracy of far-end lane detection will drop significantly for these deep learning methods for lane line detection.

And when the mainstream deep learning lane line detection algorithm is optimized, the cloud will extract a large amount of user data of vehicle terminals, which requires a large network bandwidth and involves user privacy issues

Images