基于Shi-Tomasi角点验证的线段提取算法优化方法

骆开庆; 邓军灿; 蔡伟博; 周育滨; 张健

doi:10.6054/j.jscnun.2022016

基于Shi-Tomasi角点验证的线段提取算法优化方法

doi: 10.6054/j.jscnun.2022016

华南师范大学物理与电信工程学院/电子与信息工程学院/广东省光电检测仪器工程技术研究中心，广州 510006

基金项目:

国家自然科学基金委员会-广东大数据科学中心项目 U1911401

华南师范大学大学生创新创业训练项目 202010574050

华南师范大学大学生创新创业训练项目 202110574048

详细信息

通讯作者:
张健，Email: jianzhang@m.scnu.edu.cn

中图分类号: TP391.4
计量
- 文章访问数: 270
- HTML全文浏览量: 732
- PDF下载量: 74
- 被引次数: 0
出版历程
- 收稿日期: 2021-06-12
- 刊出日期: 2022-02-25

Line Segment Extraction Algorithm Optimization Based on Shi-Tomasi Corner Detector

School of Physics and Telecommunications Engineering/School of Electronics and Information Engineering, South China Normal University/Guangdong Provincial Engineering Research Center for Optoelectronic Instrument, Guangzhou 510006, China

摘要

摘要: 针对现有的线段提取算法在图像中的天空、阴影、玻璃以及地板等模糊区域提取出较多的无意义线段的问题，提出了一种基于Shi-Tomasi角点验证的线段提取算法优化方法(ST-Lines算法)：首先，使用经典线段提取算法进行线段提取；然后，采用Shi-Tomasi角点检测算法提取角点，并利用滑动窗口对所得的角点进行非极大值抑制；最后，根据线段长度、线段端点圆形框内的角点分布情况以及K最近邻算法对每条线段进行有无意义验证，尽可能多地剔除无意义线段。并利用YorkUrban线段数据集，对ST-Lines算法与原线段提取方法进行测试对比。对比结果表明：ST-Lines算法在平均准确率、F-score、平均线段长度上有所提高，且降低了平均线段数量。
- Shi-Tomasi角点检测 /
- 无意义线段 /
- 线段提取优化 /
- K最近邻算法
Abstract: The existing line segment extraction algorithm extracts many meaningless lines from blurred areas such as the sky, shadow, glass and floor in the image. To solve this problem, a method for line segment extraction algorithm optimization based on Shi-Tomasi corner detector is proposed, which is called ST-Lines. It firstly uses a classic line segment extraction algorithm to extract lines. Then corners are extracted with the Shi-Tomasi corner detector and non-maximum suppression is applied to them with a sliding window. Finally, it verifies the meaning of each line segment and removes as many meaningless line segments as possible to optimize the line segment extraction algorithm based on the length of the line segment, the distribution of corners in the circle region of endpoints, and the K-nearest neighbor algorithm. The YorkUrban line segment dataset is used to compare ST-Lines algorithm and the original algorithm. The experimental results show that ST-Lines algorithm can improve the accuracy, F-score and the averageline length and reduce the average line number.
- Shi-Tomasi corner detector /
- meaningless line /
- line extraction optimization /
- K-nearest neighbor algorithm

HTML全文

图 1 Shi-Tomasi算法中窗口滑动时灰度值变化较大的方向

Figure 1. The gray value changes of sliding windows in each direction in the Shi-Tomasi algorithm

下载: 全尺寸图片幻灯片

图 2 通过滑动窗口对角点进行非极大值抑制

Figure 2. The non-maximum suppression of corners with the sliding window

下载: 全尺寸图片幻灯片

图 3 部分特征模糊区域的线段和角点分布

Figure 3. The distribution of line segments and corners in blurred areas

下载: 全尺寸图片幻灯片

图 4 线段验证总流程图

Figure 4. The block diagram of validation of line segments

下载: 全尺寸图片幻灯片

图 5 线段端点圆形框内角点统计方法

Figure 5. The corner statistics in circles

下载: 全尺寸图片幻灯片

图 6 KNN算法验证方法

Figure 6. The validation of the KNN algorithm

下载: 全尺寸图片幻灯片

图 7 无意义线段剔除效果

Figure 7. The result of removing meaningless line segments

下载: 全尺寸图片幻灯片

图 8 数据集中每张图片的结果对比

Figure 8. The comparison of the results of all images in the dataset

下载: 全尺寸图片幻灯片

图 9 4种算法的室内外场景结果

Figure 9. The experimental results of indoor and outdoor scenes with four algorithms

下载: 全尺寸图片幻灯片

图 10 不同λ_area 下4种算法的F-score的变化曲线

Figure 10. The change curves of F-score of four algorithms under different λ_area

下载: 全尺寸图片幻灯片

表 1 6种算法在数据集的平均结果对比

Table 1. The comparison of average results of 6 algorithms in the dataset

算法	AP	AR	F-score	NUM	LENGTH	FPS
EDLines	0.149 1	0.530 0	0.232 8	953.323 5	26.275 9	36.428 5
LSD	0.167 7	0.432 1	0.241 6	757.578 4	25.173 6	15.887 8
LSM	0.146 9	0.319 5	0.201 2	690.951 0	28.300 0	0.257 7
Linelet	0.159 1	0.507 0	0.242 2	940.656 9	23.028 7	0.037 7
EDLines+Corner	0.178 9	0.512 1	0.265 1	699.509 8	30.485 8	29.059 8
LSD+Corner	0.181 4	0.409 2	0.251 3	608.156 9	27.609 2	13.298 5

下载: 导出CSV

参考文献(21)

[1]	ETEMADI A. Robust segmentation of edge data[C]//Proceedings of the 1992 International Conference on Image Processing and its Applications. Maastricht, Netherlands: IET, 1992: 311-314.
[2]	TOPAL C, AKINLAR C. Edge drawing: a combined real-time edge and segment detector[J]. Journal of Visual Communication and Image Representation, 2012, 23(6): 862-872. doi: 10.1016/j.jvcir.2012.05.004
[3]	AKINLAR C, TOPAL C. EDLines: a real-time line segment detector with a false detection control[J]. Pattern Recognition Letters, 2011, 32(13): 1633-1642. doi: 10.1016/j.patrec.2011.06.001
[4]	BURNS J B, HANSON A R, RISEMAN E M. Extracting straight lines[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1986, 8(4): 425-455.
[5]	GROMPONE VON GIOI R, JAKUBOWICZ J, MOREL J, et al. LSD: a fast line segment detector with a false detection control[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008, 32(4): 722-732. http://www.researchgate.net/profile/Gregory_Randall/publication/41910459_LSD_A_Fast_Line_Segment_Detector_with_a_False_Detection_Control/links/559561e208ae21086d20657c.pdf
[6]	DESOLNEUX A, MOISAN L, MOREL J M. Meaningful alignments[J]. International Journal of Computer Vision, 2000, 40(1): 7-23. doi: 10.1023/A:1026593302236
[7]	HAMID N, KHAN N. LSM: perceptually accurate line segment merging[J]. Journal of Electronic Imaging, 2016, 25(6): 061620/1-11.
[8]	SUÁREZ I, MUOZ E, BUENAPOSADA J M, et al. FSG: a statistical approach to line detection via fast segments grouping[C]//Proceedings of the 2018 IEEE/RSJ International Conference on Intelligent Robots and Systems. Madrid, Spain: IEEE, 2018: 97-102.
[9]	SALAUN Y, MARLET R, MONASSE P. Multiscale line segment detector for robust and accurate SfM[C]//Proceedings of the 23rd International Conference on Pattern Recognition. Cancun, Mexico: IEEE, 2017: 2000-2005.
[10]	CHO N G, YUILLE A, LEE S W. A novel linelet-based representation for line segment detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 40(5): 1195-1208.
[11]	LIU C, ABERGEL R, GOUSSEAU Y, et al. LSDSAR, a Markovian a contrario framework for line segment detection in SAR images[J]. Pattern Recognition, 2020, 98: 107034/1-13.
[12]	LI H, YU H, YANG W, et al. ULSD: unified line segment detection across pinhole, fisheye, and spherical cameras[J]. ISPRS Journal of Photogrammetry and Remote Sen-sing, 2021, 178: 187-202. doi: 10.1016/j.isprsjprs.2021.06.004
[13]	XUE N, BAI S, WANG F, et al. Learning attraction field representation for robust line segment detection[C]//Proceedings of the 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA: IEEE, 2020: 1595-1603.
[14]	XUE N, BAI S, WANG F D, et al. Learning regional attra-ction for line segment detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(6): 1998-2013. doi: 10.1109/TPAMI.2019.2958642
[15]	MORAVEC H P. Techniques towards automatic visual obstacle avoidance[C]//Proceedings of the 1977 International Joint Conference on Artificial Intelligence. Tokyo, Japan: IJCAI, 1977: 584-584.
[16]	HARRIS C, STEPHENS M. A combined comer and edge detector[C]//Proceedings of the 4th Alvey Vision Conference. Manchester: University of Sheffield Printing Unit, 1988: 147-151.
[17]	SHI J, TOMASI C. Good features to track[C]//Procee-dings of the 1994 IEEE Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 1994: 593-600.
[18]	SMITH S M, BRADY J M. SUSAN-A new approach to low-level image processing[J]. International Journal of Computer Vision, 1997, 23(1): 45-78. doi: 10.1023/A:1007963824710
[19]	ROSTEN E, DRUMMOND T. Machine learning for high-speed corner detection[C]//Proceedings of the 8th European Conference on Computer Vision. Berlin: Springer, 2006: 430-443.
[20]	ROSTEN E, PORTER R, DRUMMOND T. Faster and be-tter: a machine learning approach to corner detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2010, 32(1): 105-119.
[21]	DENIS P, ELDER J H, ESTRADA F J. Efficient edge-based methods for estimating manhattan frames in urban imagery[C]//Proceedings of the 10th European Confe-rence on Computer Vision: Part Ⅱ. Berlin: Springer, 2008: 197-210.