使用GIoU改进非极大值抑制的目标检测算法

doi:10.12263/DZXB.20200132

电子学报 ›› 2021, Vol. 49 ›› Issue (4): 696-705.DOI: 10.12263/DZXB.20200132

使用GIoU改进非极大值抑制的目标检测算法

侯志强¹, 刘晓义¹, 余旺盛², 蒲磊², 马素刚¹, 范九伦³

1. 西安邮电大学计算机学院, 陕西西安 710121;
2. 空军工程大学信息与导航学院, 陕西西安 710077;
3. 西安邮电大学通信与信息工程学院, 陕西西安 710121

收稿日期:2020-01-20 修回日期:2020-07-06 出版日期:2021-04-25 发布日期:2021-04-25
作者简介:侯志强 男,1973年10月生,陕西省眉县人.现为西安邮电大学计算机学院教授,主要从事图像处理、计算机视觉、无人机应用以及信息融合等领域的研究;刘晓义 女,1995年8月生,陕西省澄城县人.现为西安邮电大学硕士研究生,主要从事计算机视觉、目标检测以及图像处理等领域的研究.E-mail:18829290763@163.com
基金资助:
国家自然科学基金（No.61703423，No.61473309，No.62072370）

Object Detection Algorithm for Improving Non-Maximum Suppression Using GIoU

HOU Zhi-qiang¹, LIU Xiao-yi¹, YU Wang-sheng², PU Lei², MA Su-gang¹, FAN Jiu-lun³

1. School of Computer, Xi'an University of Posts and Telecommunications, Xi'an, Shaanxi 710121, China;
2. School of Information and Navigation, Air Force Engineering University, Xi'an, Shaanxi 710077, China;
3. School of Communication and Information Engineering, Xi'an University of Posts and Telecommunications, Xi'an, Shaanxi 710121, China

Received:2020-01-20 Revised:2020-07-06 Online:2021-04-25 Published:2021-04-25

摘要/Abstract

摘要： 针对单阈值-非极大值抑制算法中出现的目标漏检和重复检测问题，本文提出了一种使用全局交并比指标GIoU（Generalized Intersection over Union）衡量目标相似度的双阈值非极大值抑制算法GDT-NMS（Generalized Dual Threshold NMS，GDT-NMS）.使用双阈值改进NMS算法和soft-NMS算法，抑制多余的检测框；在此基础上，使用GIoU替换传统的IoU计算目标间的相似度，使目标的定位更加准确；进一步，使用非线性函数赋予检测框不同比例的权值惩罚，使检测框的得分随距离呈非线性变化，目标区分度更高.改进算法在PASCAL VOC和MSCOCO上的检测精度分别为74.8%和25.9%，与使用NMS算法作为后处理的Faster R-CNN算法相比，性能分别提升了1.6%和1.5%.同时本文算法具有较快的检测速度.

关键词: 双阈值, 非极大值抑制算法, 重复检测, 后处理

Abstract: Aiming at the problem of missed detection and repeated detection in the single-threshold-non-maximum suppression algorithm,this paper proposes a dual-threshold Non-Maximum Suppression algorithm using GIoU (Generalized Intersection over Union).Using dual thresholds to improve the NMS algorithm and the soft-NMS algorithm,suppressing redundant detection boxes,not only balances the relationship between the object missed detection problem and the object false detection problem caused by the single threshold algorithm,but also reduces the occurrence of the soft-NMS algorithm.Based on the above,using GIoU instead of IoU to calculate the similarity between objects,so that the positioning of the object is more accurate;the non-linear function is used to give different weights to the proposal boxes,which makes the proposal boxes’ scores change non-linearly with distance,and the target discrimination is higher,which is more conducive to suppressing the proposal boxes.The detection accuracy of the improved algorithm on PASCAL VOC and MSCOCO is 74.8% and 25.9%,respectively.At the same time,the algorithm in this paper has a fast detection speed.

Key words: double threshold, non-maximum suppression algorithm, repeated detection, post processing

中图分类号:

TP391.4

侯志强, 刘晓义, 余旺盛, 蒲磊, 马素刚, 范九伦. 使用GIoU改进非极大值抑制的目标检测算法[J]. 电子学报, 2021, 49(4): 696-705.

HOU Zhi-qiang, LIU Xiao-yi, YU Wang-sheng, PU Lei, MA Su-gang, FAN Jiu-lun. Object Detection Algorithm for Improving Non-Maximum Suppression Using GIoU[J]. Acta Electronica Sinica, 2021, 49(4): 696-705.

参考文献

[1] Liu L,Ouyang W L,Wang X.Deep learning for generic object detection:A survey[J].arXiv Preprint,2018,arXiv:1809.02165.
[2] 黄凯奇,任伟强,谭铁牛.图像物体分类与检测算法综述[J].计算机学报,2014,37(6):1225-1240. HUANG Kai-qi,REN Wei-qiang,TAN Tie-niu.Overview of image object classification and detection algorithms[J].Chinese Journal of Computers,2014,37(6):1225-1240.(in Chinese)
[3] Shi B,Bai X,Belongie S.Detecting oriented text in natural images by linking segments[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2017.2550-2558.
[4] 周炫余,刘娟,卢笑,等.一种联合文本和图像信息的行人检测方法[J].电子学报,2017,45(1):140-146. ZHOU Xuan-yu,LIU Juan,LU Xiao,et al.A pedestrian detection method combining text and image information[J].Acta Electronica Sinica,2017,45(1):140-146.(in Chinese)
[5] 姜维,张重生,殷绪成.基于深度学习的场景文字检测综述[J].电子学报,2019,47(5):1152-1161. JIANG Wei,ZHANG Chong-sheng,YIN Xu-cheng.Summary of scene text detection based on deep learning[J].Acta Electronica Sinica,2019,47(5):1152-1161.(in Chinese)
[6] 辛鹏,许悦雷,唐红,等.全卷积网络多层特征融合的飞机快速检测[J].光学学报,2018,38(3):315003. XIN Peng,XU Yuelei,TANG Hong,et al.Rapid detection of aircraft with multi-layer feature fusion of fully convolutional networks[J].Acta Optics Sinica,2018,38(3):315003.(in Chinese)
[7] 丁鹏,张叶,贾平,等.基于视觉显著性的海面舰船检测技术[J].电子学报,2018,46(1):127-134. DING Peng,ZHANG Ye,JIA Ping,et al.Sea-based ship inspection technology basedon visual saliency[J].Acta Electronica Sinica,2018,46(1):127-134.(in Chinese)
[8] Tian Z,Huang W,He T,et al.Detecting text in natural image with connectionist text proposal network[A].European Conference on Computer Vision[C].Berlin,Cham:Springer,2016.56-72.
[9] Redmon J,Divvala S,Girshick R,et al.You only look once:Unified,real-time object detection[A].IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2016.779-788.
[10] Redmon J,Farhadi A.YOLO9000:better,faster,stronger[A].Proceedings of IEEE Conference on Computer Vision and Pattern Recognition[C].Honolulu,USA:IEEE,2017.6517-6525.
[11] Redmon Joseph,Farhadi Ali.Yolov3:An incremental improvement[J].arXiv Preprint,2018,arXiv:1804.02767.
[12] Liu W,Anguelov D,Erhan D,et al.SSD:Single shot multibox detector[A].European Conference on Computer Vision[C].Berlin,GER:Springer International Publishing,2016.21-37.
[13] Fu C Y,Liu W,Ranga A,et al.DSSD:Deconvolutional single shot detector[J].arXiv Preprint,2017,arXiv:1701.06659.
[14] Shen Z Q,Liu Z H,Li J G,et al.DSOD:Learning deeply supervised object detectors from scratch[A].Proceedings of the IEEE International Conference on Computer Vision[C].New York,USA:IEEE,2017.1919-1927.
[15] Girshick R,Donahue J,Darrell T.Rich feature hierarchies for accurate object detection and semantic segmentation[A].Proceedings of IEEE Conference on Computer Vision and Pattern Recognition[C].Columbus,USA:IEEE,2014.580-587.
[16] Girshick R.Fast R-CNN[A].Proceedings of the IEEE International Conference on Computer Vision[C].New York,USA:IEEE,2015.1440-1448.
[17] Ren S,He K M,Girshick R,et al.Faster R-CNN:Towards real-time object detection with region proposal networks[A].International Conference on Neural Information Processing Systems[C].Cambridge USA:MIT Press,2015.91-99.
[18] Henderson P,Ferrari V.End-to-end training of object class detectors for mean average precision[A].Proceedings of the 13th Asian Conference on Computer Vision[C].Taipei,China:Springer,2016.198-213.
[19] Hosang J,Benenson R,Schiele B.Learning non-maximum suppression[A].Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition[C].Honolulu,USA:IEEE,2017.6469-6477.
[20] Stewart R,Andriluka M,Ng A Y.End-to-end people detection in crowded scenes[A].Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition[C].Las Vegas,USA:IEEE,2016.2325-2333.
[21] Bodla N,Singh B,Chellappa R,et al.Soft-NMS-improving object detection with one line of code[A].IEEE International Conference on Computer Vision[C].New York,USA:IEEE,2017.5562-5570.
[22] Hamid Rezatofighi,Nathan Tsoi,JunYoung Gwak,et al.Generalized intersection over union:A metric and a loss for bounding box[J].arXiv Preprint,2019,arXiv:1902.09630.
[23] Neubeck A,Gool L V.Efficient non-maximum suppression[A].International Conference on Pattern Recognition[C].New York,USA:IEEE Computer Society,2006.850-855.
[24] Rahman M A,Wang Y.Optimizing intersection-over-union in deep neural networks for image segmentation[A].Advances in Visual Computing[M].Berlin,GER:Springer International Publishing,2016.234-244.
[25] 赵文清,严海,邵绪强.改进的非极大值抑制算法的目标检测[J].中国图象图形学报,2018,23(11):1676-1685. ZHAO Wen-qing,YAN Hai,SHAO Xu-qiang.Target detection by improved non-maximum suppression algorithm[J].Journal of Image and Graphics,2018,23(11):1676-1685.(in Chinese)
[26] 牛斌,张怡迪,马利,等.一种融合多尺度特征的多物体检测方法[J].辽宁大学学报(自然科学版),2019,46(02):118-124. NIU Bin,ZHANG Yi-di,MA Li,et al.A multi-object detection method combining multi-scale features[J].Journal of Liaoning University (Natural Science Edition),2019,46(02):118-124.(in Chinese)
[27] 张烨,许艇,冯定忠,等.基于难分样本挖掘的快速区域卷积神经网络目标检测研究[J].电子与信息学报,2019,41(06):1496-1502. ZHANG Ye,XU Ting,FENG Ding-zhong,et al.Research on target detection of fast regional convolutional neural network based on mining difficult samples[J].Journal of Electronics and Information Technology,2019,41(06):1496-1502.(in Chinese)
[28] Kong T,Sun F C,Yao A B,et al.RON:Reverse connection with objectness prior networks for object detection[A].Proceedings of IEEE Conference on Computer Vision and Pattern Recognition[C].Honolulu,USA:IEEE,2017.5244-5252.
[29] He K M,Zhang X Y,Ren S Q,et al.Deep residual learning for image recognition[A].IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2016.770-778.
[30] Huang G,Liu Z,Maaten L V D,et al.Densely connected convolutional networks[A].IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2017.2261-2269.

使用GIoU改进非极大值抑制的目标检测算法

Object Detection Algorithm for Improving Non-Maximum Suppression Using GIoU

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 6

编辑推荐

Metrics

[1]	姚桂林, 刘绍辉, 李敏, 姚鸿勋, 胡文. 后处理在数字抠像中的应用与解析[J]. 电子学报, 2017, 45(3): 719-729.
[2]	毕冉, 李建中, 高宏. 无线传感器网络中基于双阈值的分布式监测算法[J]. 电子学报, 2014, 42(8): 1594-1600.
[3]	赵德香, 马秀荣, 白媛, 程云翔. 基于阈值处理的TD-SCDMA下行多用户信道估计方法[J]. 电子学报, 2012, 40(9): 1904-1908.
[4]	汪金辉;宫娜;左磊;彭晓宏;吴武臣. 双阈值多米诺或门性能估计及其工艺浮动下的适用性分析[J]. 电子学报, 2010, 38(11): 2611-2615.
[5]	梁光明;唐朝京;刘东华;孙即祥;张志涛. 基于分割评价的多层次自适应双阈值分割算法[J]. 电子学报, 2009, 37(4): 750-752.
[6]	李牧;臧希;闫继宏;赵杰. 基于类内方差最小化及模糊控制算法的小波边缘检测技术[J]. 电子学报, 2008, 36(9): 1741-1745.