基于HSV空间的钢轨表面区域快速提取算法
|
摘要
机器视觉采集的轨道图像中包含钢轨表面区域(简称轨面区域)及干扰区域(轨道、碎石、杂草等),从复杂的轨道图像中检测缺陷目标难度大、耗时多.因此,先将轨面区域提取出来,然后对提取的轨面区域进行缺陷检测和识别可以节省大量时间.传统的钢轨表面区域提取算法多为手动设定轨面边界,自适应性较差,且对光照异常敏感.针对以上问题,就如何快速提取轨面区域进行了研究,提出了一种基于HSV空间的钢轨表面区域快速提取算法.首先将采集的RGB图像转换到HSV空间,并提取其S分量图像,以此变换来克服光照条件变化对钢轨表面区域提取带来的干扰;其次在S分量图像中绘制灰度投影曲线;然后以图像的中点为轴,将曲线分为左右两侧,分别找到左右两侧列曲线的极大值L_1、R_1和次大值L_2、R_2;最后根据极大值和次大值的关系自动确定轨面区域的边界.仿真结果表明,所研究的算法可以快速、准确地提取轨面区域,避免了手动确定轨面区域边界的问题,算法的泛化能力较好,提取精度较高(I_(oU)高达0.92),提取准确率为93.87%,提取时间较传统方法大幅降低,平均提取时间为0.046 s,为铁路线路的实时自动化检测奠定了基础.
Rail image captured by machine includes rail surface area and interferenc zone(track, gravel,weeds, etc.). It is difficult and time-consuming to detect defective targets from complex track images. Therefore, It can save a lot of time to extract the rail area first, and then detect and identify the defects in the extracted rail area.The traditional rail surface region(abbreviated as rail surface region) extraction algorithm mostly determines the rail surface boundary manually, the application range of the algorithm is narrow, poor adaptability, and sensitive to light. In order to solve the above problems, the fast extraction method of rail area is studied and an algorithm for quickly extracting rail surface region based on HSV space is proposed in the thesis. Firstly, the acquired RGB image is transformed into HSV space, and its S component image is extracted, through this transformation to overcome the interference caused by changes in lighting conditions; Secondly, a gray projection curve is drawn in the S component image; Then use the midpoint of the image as an axis to divide the curve into the left and right sides, and find the maximum and minor values of the left and right column curves respectively. Finally, the boundary of the rail surface region is automatically determined based on the relationship between the maximum value and the second largest value. The simulation results show that the proposed algorithm can accurately and quickly extract the rail surface region and avoid manually determining the boundary of the rail surface region. The generalization ability of the algorithm is better and the extraction accuracy is higher(I_(oU) is up to 0.92). the average extraction accuracy is 93.87%, and the extraction time is 0.046 s, significantly lower than the traditional method, it lays a foundation for real-time automatic detection of rail track.
Rail image captured by machine includes rail surface area and interferenc zone(track, gravel,weeds, etc.). It is difficult and time-consuming to detect defective targets from complex track images. Therefore, It can save a lot of time to extract the rail area first, and then detect and identify the defects in the extracted rail area.The traditional rail surface region(abbreviated as rail surface region) extraction algorithm mostly determines the rail surface boundary manually, the application range of the algorithm is narrow, poor adaptability, and sensitive to light. In order to solve the above problems, the fast extraction method of rail area is studied and an algorithm for quickly extracting rail surface region based on HSV space is proposed in the thesis. Firstly, the acquired RGB image is transformed into HSV space, and its S component image is extracted, through this transformation to overcome the interference caused by changes in lighting conditions; Secondly, a gray projection curve is drawn in the S component image; Then use the midpoint of the image as an axis to divide the curve into the left and right sides, and find the maximum and minor values of the left and right column curves respectively. Finally, the boundary of the rail surface region is automatically determined based on the relationship between the maximum value and the second largest value. The simulation results show that the proposed algorithm can accurately and quickly extract the rail surface region and avoid manually determining the boundary of the rail surface region. The generalization ability of the algorithm is better and the extraction accuracy is higher(I_(oU) is up to 0.92). the average extraction accuracy is 93.87%, and the extraction time is 0.046 s, significantly lower than the traditional method, it lays a foundation for real-time automatic detection of rail track.
引文
[1]蔡南平.基于图像的钢轨表面缺陷检测[D].长沙:中南大学,2014.Cai N P.Defects detection of rail surface based on image[D].Changsha:Central South University,2014.
[2]顾桂梅,黄涛.基于自适应多尺度积阈值的钢轨裂纹红外图像增强[J].铁道学报,2015,37(2):58-63.DOI:10.3969/j.issn.1001-8360.2015.02.12.Gu G M,Huang T.Infrared image enhancement of rail cracks based on adaptive multiscale product threshold[J].Journal of the Railway,2015,37(2):58-63.
[3]闵永智,殷超,党建武,等.基于图像色相值突变特征的钢轨区域快速识别方法[J].交通运输工程学报,2016,16(1):46-54.DOI:10.3969/j.issn.1671-1637.2016.01.006.Min Y Z,Yin C,Dang J W,et al.Rapid identification method of rail area based on mutation characteristics of image hue value[J].Journal of Traffic and Transportation Engineering,2016,16(1):46-54.
[4]贺振东,王耀南,毛建旭,等.基于反向PM扩散的钢轨表面缺陷视觉检测[J].自动化学报,2014,40(8):1 667-1 679.He Z D,Wang Y N,Mao J X,et al.Vision detection of rail surface defects based on reverse PM diffusion[J].Journal of Automation,2014,40(8):1 667-1 679.
[5]周鹏,徐科,张春阳,等.基于多激光线的钢轨表面缺陷在线检测方法[J].工程科学学报,2015,37(s1):18-23.Zhou P,Xu K,Zhang C Y,et al.Onlinedetection method of rail surface defects based on multilaser line[J].Journal of Engineering Science,2015,37(s1):18-23.
[6]Rowshandel H,Nicholson GL,Davis CL,et al.A combined threshold and signature match method for the automatic detection of rail RCF cracks using an ACFMsensor[C].IET Conference on Railway Condition Monitoring,Birmingham,England,2014:1-6.
[7]Vijaykumar V R,Sangamithirai S.Rail defect detection using Gabor filters with texture analysis[C].International Conference on Signal Processing,Communication and Networking,Chennai,India,2015:1-6.
[8]Ta?timur C,Karak?se M,Ak?n E,et al.Rail defect detection with real time image processing technique[C].International Conference on Industrial Informatics,Poitier,France,2017:411-415.
[9]邬锋,茅正冲.铁轨表面缺陷图像增强与分割算法[J].计算机仿真,2015,32(10):159-162.DOI:10.3969/j.issn.1006-9348.2015.10.035.Wu F,Mao Z C.Image enhancement and segmentation algorithm for rail surface defects[J].Computer Simulation,2015,32(10):159-162.
[10]唐湘娜,王耀南.铁轨表面缺陷的视觉检测与识别算法[J].计算机工程,2013,39(3):25-30.DOI:10.3969/j.issn.1000-3428.2013.03.006.Tang X N,Wang Y N.Vision detectionand identification algorithm for rail surface defects[J].Computer Engineering,2013,39(3):25-30.
[11]程天栋.轨检图像中钢轨表面缺陷分析与识别[D].兰州:兰州交通大学,2017.Cheng T D.Analysis and identificationof rail surface defects in track inspection images[D].Lanzhou:Lanzhou Jiaotong University,2017.
[12]贺振东,王耀南,刘洁,等.基于背景差分的高铁钢轨表面缺陷图像分割[J].仪器仪表学报,2016,37(3):640-649.DOI:10.3969/j.issn.0254-3087.2016.03.022.He Z D,Wang Y N,Liu J,et al.Image segmentation of rail surface defects onhighspeed railway based on background difference[J].Journal of Instrumentation,2016,37(3):640-649.
[13]闵永智,岳彪,马宏锋,等.钢轨表面缺陷图像自适应分割算法[J].北京工业大学学报,2017,43(10):1 472-1 479.DOI:10.11936/bjutxb2016100026.Min Y Z,Yue B,Ma H F,et al.Adaptive segmentation algorithm for rail surface defects image[J].Journal of Beijing Polytechnic University,2017,43(10):1 472-1 479.
[14]王时丽.基于机器视觉的钢轨表面缺陷检测技术研究[D].绵阳:西南科技大学,2016.Wang S L.Study on the detection technology of rail surface defects based on machine vision[D].Mianyang:Southwest University of Science and Technology,2016.
[15]夏利民,蔡南平,杨宝娟.基于图像差异和视觉反差的钢轨表面缺陷检测[J].计算机工程与设计,2014,35(6):2 052-2 055.DOI:10.3969/j.issn.1000-7024.2014.06.034.Xia L M,Cai N P,Yang B J.Detection of rail surface defects based on image difference and visual contrast[J].Computer Engineering and Design,2014,35(6):2 052-2 055.
[16]赵宏宇,肖创柏,禹晶,等.马尔科夫随机场模型下的Retinex夜间彩色图像增强[J].光学精密工程,2014,22(4):1 048-1 055.Zhao H Y,Xiao C B,Yu J,et al.Retinex night color image enhancement under Markov random field model[J].Optical Precision Engineering,2014,22(4):1 048-1 055.
[17]Girshick R.Fast RCNN[C].IEEE International Conference on Computer Vision,Santiago,Chile,2015:1 440-1 448.
[2]顾桂梅,黄涛.基于自适应多尺度积阈值的钢轨裂纹红外图像增强[J].铁道学报,2015,37(2):58-63.DOI:10.3969/j.issn.1001-8360.2015.02.12.Gu G M,Huang T.Infrared image enhancement of rail cracks based on adaptive multiscale product threshold[J].Journal of the Railway,2015,37(2):58-63.
[3]闵永智,殷超,党建武,等.基于图像色相值突变特征的钢轨区域快速识别方法[J].交通运输工程学报,2016,16(1):46-54.DOI:10.3969/j.issn.1671-1637.2016.01.006.Min Y Z,Yin C,Dang J W,et al.Rapid identification method of rail area based on mutation characteristics of image hue value[J].Journal of Traffic and Transportation Engineering,2016,16(1):46-54.
[4]贺振东,王耀南,毛建旭,等.基于反向PM扩散的钢轨表面缺陷视觉检测[J].自动化学报,2014,40(8):1 667-1 679.He Z D,Wang Y N,Mao J X,et al.Vision detection of rail surface defects based on reverse PM diffusion[J].Journal of Automation,2014,40(8):1 667-1 679.
[5]周鹏,徐科,张春阳,等.基于多激光线的钢轨表面缺陷在线检测方法[J].工程科学学报,2015,37(s1):18-23.Zhou P,Xu K,Zhang C Y,et al.Onlinedetection method of rail surface defects based on multilaser line[J].Journal of Engineering Science,2015,37(s1):18-23.
[6]Rowshandel H,Nicholson GL,Davis CL,et al.A combined threshold and signature match method for the automatic detection of rail RCF cracks using an ACFMsensor[C].IET Conference on Railway Condition Monitoring,Birmingham,England,2014:1-6.
[7]Vijaykumar V R,Sangamithirai S.Rail defect detection using Gabor filters with texture analysis[C].International Conference on Signal Processing,Communication and Networking,Chennai,India,2015:1-6.
[8]Ta?timur C,Karak?se M,Ak?n E,et al.Rail defect detection with real time image processing technique[C].International Conference on Industrial Informatics,Poitier,France,2017:411-415.
[9]邬锋,茅正冲.铁轨表面缺陷图像增强与分割算法[J].计算机仿真,2015,32(10):159-162.DOI:10.3969/j.issn.1006-9348.2015.10.035.Wu F,Mao Z C.Image enhancement and segmentation algorithm for rail surface defects[J].Computer Simulation,2015,32(10):159-162.
[10]唐湘娜,王耀南.铁轨表面缺陷的视觉检测与识别算法[J].计算机工程,2013,39(3):25-30.DOI:10.3969/j.issn.1000-3428.2013.03.006.Tang X N,Wang Y N.Vision detectionand identification algorithm for rail surface defects[J].Computer Engineering,2013,39(3):25-30.
[11]程天栋.轨检图像中钢轨表面缺陷分析与识别[D].兰州:兰州交通大学,2017.Cheng T D.Analysis and identificationof rail surface defects in track inspection images[D].Lanzhou:Lanzhou Jiaotong University,2017.
[12]贺振东,王耀南,刘洁,等.基于背景差分的高铁钢轨表面缺陷图像分割[J].仪器仪表学报,2016,37(3):640-649.DOI:10.3969/j.issn.0254-3087.2016.03.022.He Z D,Wang Y N,Liu J,et al.Image segmentation of rail surface defects onhighspeed railway based on background difference[J].Journal of Instrumentation,2016,37(3):640-649.
[13]闵永智,岳彪,马宏锋,等.钢轨表面缺陷图像自适应分割算法[J].北京工业大学学报,2017,43(10):1 472-1 479.DOI:10.11936/bjutxb2016100026.Min Y Z,Yue B,Ma H F,et al.Adaptive segmentation algorithm for rail surface defects image[J].Journal of Beijing Polytechnic University,2017,43(10):1 472-1 479.
[14]王时丽.基于机器视觉的钢轨表面缺陷检测技术研究[D].绵阳:西南科技大学,2016.Wang S L.Study on the detection technology of rail surface defects based on machine vision[D].Mianyang:Southwest University of Science and Technology,2016.
[15]夏利民,蔡南平,杨宝娟.基于图像差异和视觉反差的钢轨表面缺陷检测[J].计算机工程与设计,2014,35(6):2 052-2 055.DOI:10.3969/j.issn.1000-7024.2014.06.034.Xia L M,Cai N P,Yang B J.Detection of rail surface defects based on image difference and visual contrast[J].Computer Engineering and Design,2014,35(6):2 052-2 055.
[16]赵宏宇,肖创柏,禹晶,等.马尔科夫随机场模型下的Retinex夜间彩色图像增强[J].光学精密工程,2014,22(4):1 048-1 055.Zhao H Y,Xiao C B,Yu J,et al.Retinex night color image enhancement under Markov random field model[J].Optical Precision Engineering,2014,22(4):1 048-1 055.
[17]Girshick R.Fast RCNN[C].IEEE International Conference on Computer Vision,Santiago,Chile,2015:1 440-1 448.