交通场景中改进SSD算法的小尺度行人检测研究

doi:10.3778/j.issn.1002-8331.2008-0452

摘要/Abstract

摘要： 由于交通场景中的行人目标所处的背景环境复杂、目标较小等因素，使得目前的行人检测算法在实际应用中存在检测精度不高、检测速度较慢的问题。因此行人检测模块作为高级辅助驾驶系统的核心模块，一直以来都是目标检测的研究热点之一。针对交通场景中小尺度行人目标，将传统的SSD网络结构中的主干网络卷积层结合Inception模块中的稀疏连接来优化卷积结构，从而增强网络的特征提取能力。同时利用残差结构组成的预测模块代替传统的两个3×3大小的卷积核来进一步提取特征图的深层特征，提高对小尺度行人目标的检测精度。引入Focal Loss函数作为网络的分类损失函数，使得损失函数更加关注于包含更多有用信息的困难负样本，解决训练过程中正负样本不平衡的问题，加快网络的收敛和稳定。实验结果表明，对于交通场景中小尺度的行人目标改进的SSD算法在检测精度和速度上都有所提高。

关键词: 行人检测, SSD算法, 残差块, Focal Loss函数

Abstract: Due to the complex background environment and small-scale pedestrian target in traffic scene, pedestrian detection algorithm has the problems of low detection accuracy and slow detection speed in practical application. So as the core module of advanced driving assistance system, pedestrian detection has always been one of hot-spots of target detection. Aiming at the small-scale pedestrian targets in traffic scenes and optimizing the convolution structure, the convolution layer of the backbone network in the traditional SSD network is combined with the sparse connection of inception module, so as to enhance the feature extraction ability of the network. At the same time, to further extract the deep features of the feature map, residual structure replaces traditional 3×3 size of convolution kernels in the prediction module, which?improves detection accuracy of small-scale pedestrian targets. Finally, the Focal Loss function is?taken?as the classification loss function of network, which makes the loss function pay more attention to the difficult negative samples, solves the problem of unbalanced positive and negative samples in the training process, and accelerates the convergence and stability of the network. The experimental results show that the improved SSD algorithm can improve the detection accuracy and speed for small-scale pedestrian targets in traffic scenes.

Key words: pedestrian detection, SSD algorithm, residual block, Focal Loss function

汪慧兰, 戴舒, 刘丹, 王桂丽. 交通场景中改进SSD算法的小尺度行人检测研究[J]. 计算机工程与应用, 2022, 58(2): 201-207.

WANG Huilan, DAI Shu, LIU Dan, WANG Guili. Small-Scale Pedestrian Detection in Traffic Scenes Based on Improved SSD Algorithm[J]. Computer Engineering and Applications, 2022, 58(2): 201-207.

参考文献

[1] 张顺，龚怡宏，王进军.深度卷积神经网络的发展及其在计算机视觉领域的应用[J].计算机学报，2019，42（3）：453-482.
ZHANG S，GONG Y H，WANG J J.The development of deep convolution neural network and its applications on computer vision[J].Chinese Journal of Computers，2019，42（3）：453-482.
[2] 苏松志，李绍滋，陈淑媛，等.行人检测技术综述[J].电子学报，2012，40（4）：814-820.
SU S Z，LI S Z，CHEN S Y.Overview of pedestrian detection technology[J].Acta Electronica Sinica，2012，40（4）：814-820.
[3] TIAN Z，SHEN C，CHEN H，et al.FCOS：fully convolutional one-stage object detection[C]//2019 IEEE/CVF International Conference on Computer Vision（ICCV），2019：9626-9635.
[4] 曾接贤，程潇.结合单双行人DPM模型的交通场景行人检测[J].电子学报，2016，44（11）：2668-2675.
ZENG J X，CHEN X.Pedestrian detection combined with single and couple pedestrian DPM models in traffic scene[J].Acta Electronica Sinica，2016，44（11）：2668-2675.
[5] KRIZHEVSKY A，SUTSKEVER I，HINTON G E.ImageNet classification with deep convolutional neural networks[C]//Proceedings of the 25th International Conference on Neural Information Processing Systems，Lake Tahoe，Nevada，USA，2012：1097-1105.
[6] WEN Z，DU Y H，YOSHIDA T，et al.DRI-RCNN：an approach to deceptive review identification using recurrent convolutional neural network[J].Information Processing ＆ Management，2018，54（4）：576-592.
[7] GIRSHICK R.Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision（ICCV），Santiago，2015：1440-1448.
[8] REN S，HE K，GIRSHICK R，et al.Faster R-CNN：towards real-time object detection with region proposal networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence，2017，39（6）：1137-1149.
[9] REDMON J，DIVVALA S，GIRSHICK R，et al.You only look once：unified，real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition（CVPR），Las Vegas，NV，2016：779-788.
[10] LIU W，ANGUELOV D，ERHAN D，et al.SSD：single shot multibox detector[C]//European Conference on Computer Vision，2016：563-590.
[11] FU C Y，LIU W，RANGA A，et al.DSSD：deconvolutional single shot detector[C]//IEEE International Conference on Computer Vision and Pattern Recognition，2017：910-925.
[12] LI Z，ZHOU F.FSSD：feature fusion single shot multibox detector[J].arXiv：1512.02325，2015.
[13] JEONG J，?PARK H，?KWAK N.Enhancement of SSD by concatenating feature maps for object detection[J].arXiv：1705.09587，2017.
[14] LIN T Y，GOYAL P，GIRSHICK R，et al.Focal loss for dense object detection[J].IEEE Transactions on Pattern Analysis & Machine Intelligence，2020，42（2）：318-327.
[15] NEUBECK A，VAN GOOL L J.Efficient non-maximum suppression[C]//18th International Conference on Pattern Recognition（ICPR 2006），Hong Kong，China，2006：850-855.
[16] LIN T Y，DOLLáR P，GIRSHICK R，et al.Feature pyramid networks for object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition（CVPR），Honolulu，HI，2017：936-944.
[17] SZEGEDY C，IOFFE S，VANHOUCKE V.Inception-v4，Inception-ResNet and the impact of residual connections on learning[J].arXiv：1602.07261，2016.
[18] HE K，ZHANG X，REN S，et al.Deep residual learning for image recognition[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas，USA：IEEE，2016：770-778.
[19] WOJEK C，DOLLAR P，SCHIELE B.Pedestrian detection：an evaluation of the state of the art[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2012，34（4）：743-761.
[20] REDMON J，FARHADI A.Yolov3：an incremental improvement[J].arXiv：1804.02767，2018.
[21] BOCHKOVSKIY A，WANG C Y，LIAO H Y M.YOLOv4：optimal speed and accuracy of object detection[J].arXiv：2004.10934，2020.