一种采用区域知识蒸馏网络的车道线检测算法  

Lane Detection Algorithm Based on Regional Knowledge Distillation Network

作　　者：叶飞 刘子龙[1] YE Fe;LIU Zi-long(School of Optical-Electrical and Computer Engineering,University of Shanghai and Technology,Shanghai 200082,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200082

出　　处：《小型微型计算机系统》2022年第11期2348-2353,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61603255)资助。

摘　　要：目前,深度学习技术被广泛应用于车道线检测,但是弱光条件下检测至今仍是一个挑战.主要原因有两个:第一,弱光数据不足;第二,模型鲁棒性不强.针对这两个问题,本文提出了一种改进循环生成对抗网络数据增强的方法来解决弱光数据不足的情况,避免了人工增加数据的复杂度;另外根据图片中车道线与背景区域之间的联系,使用一种区域亲和知识蒸馏的方法,对检测模型性能进行优化,提高模型对图片各个区域之间特征理解,提高模型的检测精度.与目前主流的车道线检测算法进行实验对比,本文提出的车道线检测算法对弱光环境的检测速度快,精度更高,不同环境的鲁棒性更强.At present,deep learning technology is widely used in lane detection,but detection under weak light conditions is still a challenge.There are two main reasons:first,the weak light data is insufficient;second,the robustness of the model is not strong.In order to solve these two problems,this paper propose a method to improve the circulation generation network data enhancement to solve the problem of insufficient weak light data,and avoid artificially increasing the complexity of data.In addition,according to the relationship between the image lane line and the background area,a region affinity knowledge distillation method was used to optimize the performance of the detection model,and improve the performance of the model for each image.Compared with the current mainstream lane detection algorithms,the lane line detection algorithm proposed in this paper has faster detection speed,higher accuracy and stronger robustness in different environments.

关 键 词：车道线检测 弱光条件 循环生成对抗网络 知识蒸馏 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]