中圖分類號(hào)： U495；TP391.41 文獻(xiàn)標(biāo)識(shí)碼： A DOI： 10.19358/j.issn.2096-5133.2022.06.010 引用格式： 陸智臣，吳麗君，陳志聰，等. 基于方向一致?lián)p失的輕量車道線檢測(cè)[J].信息技術(shù)與網(wǎng)絡(luò)安全，2022，41(6)：57-63,72.
Lightweight lane line detection based on directional consistency loss
Lu Zhichen，Wu Lijun，Chen Zhicong，Lin Peijie，Cheng Shuying
(School of Physics and Information Engineering，F(xiàn)uzhou University，F(xiàn)uzhou 350108，China)
Abstract： At present, the lightweight lane line detection network has problems such as poor curve detection effect, insufficient network receptive field and limited real-time performance.Therefore, this paper proposes an improved lightweight lane detection network model. Firstly, to improve the curve detection effect, a direction consistency loss is designed to make the model suitable for the curve scene.Secondly, in order to improve the real-time performance of the network while improving the receptive field, a fusion network of self-attention mechanism and RepVGG is proposed as the backbone network of the model. The total F1-measure of the model tested on the CULane test set reached 70.7%, the accuracy of the test on the Tusimple test set reached 95.92%, and its average inference speed reached 408 FPS. The experimental results show that the model has a certain improvement in performance compared with the current lightweight model, especially the lane line detection effect in the curve scene is significantly improved.
Key words : deep learning；lane detection；directional consistency loss；lightweight network
