变光照下基于改进的deeplabv3+的车道线检测方法  

Lane Detection Method Based on Improved Deeplabv3+under Variable Illumination

作　　者：陈涵露 宋小军 CHEN Han-lu;SONG Xiao-jun(Shanghai University of Electric Power,College of Electronics and Information Engineering,Shanghai 200120,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海200120

出　　处：《计算机仿真》2024年第11期178-181,452,共5页Computer Simulation

摘　　要：针对因变光照环境下车道线特征难提取、目标难捕捉,导致的检测误差大的问题,提出一种基于改进deeplabv3+的车道线检测方法。采用小线段形式建立车道线数学模型,利用线段拟合方法计算左、右两侧车道线上各点间的对应分布距离差值,通过差值比对获取两侧各点的对应关系,定义阈值约束检测目标的最大范围。采用改进deeplabv3+算法建立图像网格,网格四个边界置信度相等。获取车道线检测最佳阈值,将光照环境下车道线变换参数作为对比空间阈值,实行目标数据比对,并通过寻找峰值来得到精准度最高的检测数值,完成车道线检测。实验结果证明,所提方法的检测精准度高、误检率低,能适应不同的变光照环境。Aiming at the problem of large detection errors caused by the difficulty of lane feature extraction and target capture in variable illumination,this paper put forward a method of detecting lane lines based on improved deeplabv3+.At first,a mathematical model of the lane was constructed in the form of small line segments.Correspondingly,the difference of distribution distance between points on the left and right lanes was calculated by the line fiing method.Then,the corresponding relationship of points on both sides was obtained after the difference comparison.Meanwhile,the maximum range of detection targets based on the threshold constraint was determined.Moreover,the image grid was built by the improved deeplabv3+algorithm.The four boundaries of the grid should have the same confidence.After that,the best threshold for lane detection was calculated,and the lane transformation parameters under the illumination environment were used as the contrast space threshold.Furthermore,the comparison between target data was carried out.Finally,the detection value with the highest accuracy was obtained by searching for the peak value.Thus,the lane detection was completed.Experimental results show that the proposed method has high detection accuracy and low false detection rate,and can adapt to different variable illumination.

关 键 词：变光照环境车道线检测 图像网格 置信度 

分 类 号：TP123[自动化与计算机技术—控制理论与控制工程]