基于Anchor-free的交通标志检测

doi:10.12082/dqxxkx.2020.190424

摘要/Abstract

摘要：

交通标志检测是自动驾驶中的重要研究方向,实时准确地从街景图像中检测交通标志对实现自动驾驶及智慧城市的发展具有重要意义。传统的算法基于颜色、形状特征进行检测,只能提取特定种类的交通标志,算法无法同时检测不同类型的交通标志。基于图像特征+机器学习分类器的算法需要人工设计特征,算法速度较慢。主流的基于深度学习的方法多基于先验框,在网络设计上引入了额外的超参数,且在训练过程中产生过量的冗余边界框,容易造成正负样本不平衡。本文受Anchor-free思想的启发,引用YOLO检测器直接回归物体边界框的思路,提出一种基于Anchor-free的实时交通标志检测网络AF-TSD（Anchor-free Traffic Sign Detection）。AF-TSD摒弃了先验框的设计,并引入自适应采样位置可变卷积与注意力机制,大大提高网络的特征表达能力。本文开展大量对比实验,实验结果表明本文提出的AF-TSD交通标志检测网络速度接近主流算法,但精度优于主流算法,在德国GTSDB交通标志检测数据集上取得了96.80%的精度,检测速度平均单张图片32 ms,达到实时检测的要求。

关键词: 众源地理信息数据, 交通标志检测, 卷积神经网络, 可变形卷积, 注意力机制, Anchor-free, AF-TSD

Abstract:

Traffic signs are essential elements in High Definition (HD) maps and hence very important for vehicles in autonomous driving. Real-time and accurate detection of traffic signs from street level images is of great significance for the development of autonomous driving. Conventional algorithms detect traffic signs based on image color and shape features, and can only work for specific kinds of traffic signs. Algorithms based on image feature and machine learning classifier need artificial designed features, and the detection speed is slow. To date, many approaches using deep learning methods have been developed based on anchor boxes, which introduce extra hyper parameters in network design. When switching to a different detection task, anchor boxes need to be redesigned. Anchor-based methods also generate massive redundant anchor boxes during model training, which easily cause imbalance between positive and negative samples. Inspired by the idea of anchor-free and YOLO, this paper proposed a real-time traffic sign detection network called AF-TSD, which regresses object boundary directly. AF-TSD adopts an effective convolution module named deformable convolution to enhance the feature expression ability of convolutional neural networks. This module adds 2D offsets to the regular grid sampling locations in the standard convolution. It also modulates input feature amplitudes from different spatial locations/bins. Both the offsets and amplitudes are learned from the preceding feature maps, via additional convolutional layers. In addition, AF-TSD introduces attention mechanism. It is inserted after fusion of the feature pyramid, and adaptively recalibrates channel-wise feature responses by explicitly modeling the interdependencies between channels. This module first squeezes global spatial information into a channel descriptor. Then the excitation operator maps the input-specific descriptor to a set of channel weights. The attention mechanism in this paper is lightweight and imposes only a slight increase in model complexity and computational burden. To test the superiority of AF-TSD, extensive comparative experiments were carried out. We first evaluated the influence of different modules on detection precision. The experimental results show that the deformable convolution and attention mechanism can help extract features of traffic signs. Then, AF-TSD was compared with mainstream detection networks, including Faster R-CNN, RetinaNet, and YOLOv3. Our proposed AF-TSD traffic sign detection network achieved 96.80% of mAP on GTSDB traffic sign detection dataset, which was superior to mainstream detection algorithms. The average detection speed was 32ms per images, which can meet the requirements of real-time detection.

Key words: VGI data, traffic sign detection, convolutional neural networks, deformable convolution, attention mechanism, anchor-free, AF-TSD

范红超, 李万志, 章超权. 基于Anchor-free的交通标志检测[J]. 地球信息科学学报, 2020, 22(1): 88-99.DOI:10.12082/dqxxkx.2020.190424

FAN Hongchao, LI Wanzhi, ZHANG Chaoquan. Anchor-Free Traffic Sign Detection[J]. Journal of Geo-information Science, 2020, 22(1): 88-99.DOI:10.12082/dqxxkx.2020.190424

图/表 11

图1

图2

图3

图7

图4

图5

图6

图8

表1

AF-TSD网络结构设计对比"

基础网络	输入图像尺寸	mAP/%
VGG	$608 × 608$	95.40
VGG-DCN	$608 × 608$	96.29
VGG-DCN-Attention	$608 × 608$	96.80
VGG-Attention	$608 × 608$	96.02

表1

表2

AF-TSD与Faster R-CNN、RetinaNet、YOLOv3及YOLOv3（Anchor-free）之间的性能对比"

方法	输入图像尺寸像元×像元	mAP/%	s/每张图
Faster R-CNN	$608 × 608$	88.50	0.120
RetinaNet	$608 × 608$	92.43	0.094
YOLOv3	$608 × 608$	93.54	0.024
YOLOv3(Anchor-free)	$608 × 608$	94.92	0.026
AF-TSD	$608 × 608$	96.80	0.032

表2

图9

参考文献 41

[1]	De l E A, Moreno L E, Salichs M A , et al. Road traffic sign detection and classification[J]. IEEE Transactions on Industrial Electronics, 1997,44(6):848-859.
[2]	Ellahyani A, El Ansari M, El Jaafari I , et al. Traffic sign detection and recognition using features combination and random forests[J]. International Journal of Advanced Computer Science and Applications, 2016,7(1):683-693.
[3]	Miura J, Kanda T, Shirai Y . An active vision system for real-time traffic sign recognition[C]// Intelligent Transportation Systems, 2000. Proceedings, IEEE, 2000: 52-57.
[4]	徐迪红, 唐炉亮 . 基于颜色和标志边缘特征的交通标志检测[J]. 武汉大学学报·信息科学版, 2008,33(4):433-436.
	[ Xu D H, Tang L L . A pyramid-based cracks statistical model fot massive pavement images[J]. Geomatics and Information Science of Wuhan University, 2008,33(4):433-436. ]
[5]	张静, 何明一, 戴玉超 , 等. 多特征融合的圆形交通标志检测[J]. 模式识别与人工智能, 2011,24(2):226-232.
	[ Zhang J, He M Y, Dai Y C , et al. Mutil-feature fusion based circular traffic sigh detection[J]. Patten Recognition and Artifitial Intelligence, 2011,24(2):226-232. ]
[6]	贾永红, 胡志雄, 周明婷 , 等. 自然场景下三角形交通标志的检测与识别[J]. 应用科学学报, 2014,32(4):423-426.
	[ Jia Y H, Hu Z X, Zhou M T , et al. Detection and recognition of triangular traffic signs in natural scenes[J]. Journal of Applied Sciences, 2014,32(4):423-426. ]
[7]	Viola P, Jones M . Rapid object detection using a boosted cascade of simple features[J]. Computer Vision and Pattern Recognition, 2001,1(511-518):3.
[8]	Jiao J, Zhong Z, Park J , et al. A robust multi-class traffic sign detection and classification system using asymmetric and symmetric features[C]// IEEE International Conference on Systems, Man and Cybernetics. IEEE Press, 2009: 3421-3427.
[9]	Liu C, Chang F, Chen Z . Rapid multiclass traffic sign detection in high-resolution images[J]. IEEE Transactions on Intelligent Transportation Systems, 2014,15(6):2394-2403.
[10]	Dalal N, Triggs B . Histograms of oriented gradients for human detection[C]// Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on. IEEE, 2005: 886-893.
[11]	Xie Y, Liu L F, Li C H , et al. Unifying visual saliency with HOG feature learning for traffic sign detection[J]. Intelligent Vehicles Symposium IEEE, 2009: 24-29.
[12]	Girshick R, Donahue J, Darrell T , et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2014: 580-587.
[13]	Ren S, He K, Girshick R , et al. Faster r-cnn: Towards real-time object detection with region proposal networks[C]// Advances in neural information processing systems, 2015: 91-99.
[14]	Dai J, Li Y, He K , et al. R-fcn: Object detection via region-based fully convolutional networks[C]// Advances in neural information processing systems, 2016: 379-387.
[15]	Cai Z, Vasconcelos N . Cascade r-cnn: Delving into high quality object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2018: 6154-6162.
[16]	Huang L, Yang Y, Deng Y , et al. DenseBox: Unifying landmark localization with end to end object detection[J]. Computer Science, 2015(2):12-19.
[17]	Liu W, Anguelov D, Erhan D , et al. Ssd: Single shot multibox detector[C]// European conference on computer vision. Springer, Cham, 2016: 21-37.
[18]	Lin T Y, Goyal P, Girshick R , et al. Focal loss for dense object detection[C]// Proceedings of the IEEE international conference on computer vision, 2017: 2980-2988.
[19]	Lin T Y, Maire M, Belongie S , et al. Microsoft coco: Common objects in context[C]// European conference on computer vision. Springer, Cham, 2014: 740-755.
[20]	Yang T, Zhang X, Li Z , et al. Metaanchor: Learning to detect objects with customized anchors[C]// Advances in Neural Information Processing Systems. 2018: 320-330.
[21]	Law H, Deng J . Cornernet: Detecting objects as paired keypoints[C]// Proceedings of the European Conference on Computer Vision (ECCV), 2018: 734-750.
[22]	Zhou X, Zhuo J, Krahenbuhl P . Bottom-up object detection by grouping extreme and center points[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 850-859.
[23]	Huang L, Yang Y, Deng Y , et al. Densebox: Unifying landmark localization with end to end object detection[J]. arXiv preprint arXiv:1509.04874, 2015.
[24]	Yu J, Jiang Y, Wang Z , et al. Unitbox: An advanced object detection network[C]// Proceedings of the 24th ACM international conference on Multimedia. ACM, 2016: 516-520.
[25]	Tian Z, Shen C, Chen H , et al. FCOS: Fully Convolutional One-Stage Object Detection[J]. arXiv preprint arXiv:1904.01355, 2019.
[26]	Kong T, Sun F, Liu H , et al. FoveaBox: Beyond anchor-based object detector[J]. arXiv preprint arXiv:1904.03797, 2019.
[27]	Redmon J, Divvala S, Girshick R , et al. You only look once: Unified, real-time object detection[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 779-788.
[28]	Simonyan K, Zisserman A . Very deep convolutional networks for large-scale image recognition[J]. Computer Science, 2014(2):34-46.
[29]	Dai J, Qi H, Xiong Y , et al. Deformable convolutional networks[C]// Proceedings of the IEEE international conference on computer vision. 2017: 764-773.
[30]	Zhu X, Hu H, Lin S , et al. Deformable convnets v2: More deformable, better results[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2019: 9308-9316.
[31]	Lin T Y, Dollár P, Girshick R , et al. Feature pyramid networks for object detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 2117-2125.
[32]	Hu J, Shen L, Sun G . Squeeze-and-excitation networks[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2018: 7132-7141.
[33]	http://benchmark.ini.rub.de/?section=gtsdb&subsection=news
[34]	He K, Gkioxari G, Dollár P , et al. Mask r-cnn[C]// Computer Vision (ICCV), 2017 IEEE International Conference on. IEEE, 2017: 2980-2988.
[35]	Girshick R . Fast r-cnn[C]// Proceedings of the IEEE international conference on computer vision. 2015: 1440-1448.
[36]	Fu C Y, Liu W, Ranga A , et al. Dssd: Deconvolutional single shot detector[J]. arXiv preprint arXiv:1701.06659, 2017.
[37]	Szegedy C, Liu W, Jia Y , et al. Going deeper with convolutions[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9.
[38]	Ioffe S, Szegedy C . Batch normalization: Accelerating deep network training by reducing internal covariate shift[C]// International Conference on International Conference on Machine Learning, 2015: 423-434.
[39]	Szegedy C, Vanhoucke V, Ioffe S , et al. Rethinking the inception architecture for computer vision[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 2818-2826.
[40]	Szegedy C, Ioffe S, Vanhoucke V , et al. Inception-v4, inception-resnet and the impact of residual connections on learning[C]// Thirty-First AAAI Conference on Artificial Intelligence, 2017.
[41]	Ma N, Zhang X, Zheng H T , et al. Shufflenet v2: Practical guidelines for efficient cnn architecture design[C]// Proceedings of the European Conference on Computer Vision (ECCV), 2018: 116-131.