复杂场景下隧道电缆图像分割算法

An Algorithm for Tunnel Cable Image Segmentation under Complex Environment

目的分析图像空间分布和灰度分布特征,改进区域生长图像分割方法,解决光照不均,墙面多种不利因素影响造成的电缆图像分割耗时长、效果差的问题。方法首先按照墙面不利情况对图像进行分类,采用灰度均值方向投影法分析各类图像灰度分布特性,利用包络拟合离差获取电缆ROI,结合ROI空间分布信息,进行种子点初始化和终止准则设定,大大降低待处理数据量,同时避开光照不均和墙面不利因素的影响,并与K-Means聚类、全局区域生长、Unet语义分割等方法进行对比。结果对于大小为1000×1800的图像,文中方法平均分割时间为0.42 s,对于各类数据集,最大误检率和漏检率只有4%。结论文中方法有效克服了区域生长分割效果差、耗时长的缺陷,能同时解决光照不均和各种墙面不利因素影响下电缆准确分割的问题,分割效果好、耗时少。

The work aims to analyze the characteristics of image spatial distribution and gray distribution,improve the region growing segmentation method,and solve the problems of long time consumption and poor effect of cable image segmentation caused by uneven illumination and various adverse factors on the tunnel wall.Firstly,the images were classified according to the adverse conditions of the wall.Then,the gray distribution characteristics of various images were analyzed by means of vertical projection of mean gray.After that,the cable ROI was obtained with the help of the envelope fitting deviation,and the seed point initialization and termination criteria were set in combination with the spatial distribution information of ROI,so as to greatly reduce the amount of data to be processed and avoid the influence of uneven illumination and adverse factors of the wall.Finally,comparisons were made with K-means clustering,global region growth and Unet semantic segmentation.For images of 1000×1800,the average segmentation time was about 0.42 s.For all kinds of adverse situations,the maximum false detection rate and missed detection rate were only 4%.The proposed method effectively overcomes the drawbacks of poor segmentation effect and longtime consumption of region growing method,and efficiently and successfully realizes accurate cable segmentation under the influence of non-uniform illumination and various adverse factors.