面向车辆和行人检测的KM-SSD方法

计算机与现代化 ›› 2021, Vol. 0 ›› Issue (03): 51-56.

面向车辆和行人检测的KM-SSD方法

（1.江西科技师范大学通信与电子学院,江西南昌330000; 2.麦格纳汽车系统（苏州）有限公司,江苏苏州215026）

出版日期:2020-03-30 发布日期:2021-03-24
作者简介:郑钦浩（1996—）,男,广东饶平人,硕士研究生,研究方向:模式识别与智能系统,目标检测,E-mail: qinhao.zheng@aliyun.com; 杨贞（1985—）,男,山东菏泽人,讲师,博士,研究方向:机器学习,深度学习,图像处理,目标检测与跟踪,E-mail: yangzhen@jxstnu.edu.cn; 杨振（1985—）,男,山东菏泽人,硕士,研究方向:车辆检测和故障诊断,E-mail: Zhen.Yang@magna.com。
基金资助:
国家自然科学基金资助项目（61866016）; 江西科技师范大学青年拔尖项目（2018QNBJRC002）; 博士科研启动基金资助项目（2017BSQD015）; 校级自然科学重点培育基金资助项目（2017ZDPYJD005）

KM-SSD Method for Vehicle and Pedestrian Detection

（1. College of Communication and Electronics, Jiangxi Science & Technology Normal University, Nanchang 330000, China;
2. Magna Automotive Parts （Suzhou） Co. Ltd., Suzhou 215026, China）

Online:2020-03-30 Published:2021-03-24

摘要/Abstract

摘要： 传统SSD改进方法在提升SSD目标检测精度的同时会降低其检测速度。针对这一问题，本文以SSD为基础，提出改进的KM-SSD方法。该方法首先利用K-means+〖KG-*3〗+聚类算法自适应学习先验框的宽高比例；然后设计高效特征融合模块实现高低层特征信息融合；最后本文在具有挑战性的KITTI数据集上对KM-SSD方法进行验证。实验结果表明，SSD的mAP为62.7%，平均检测时间为0.162 s；KM-SSD的mAP为69.8%，平均检测时间为0.133 s。因此，KM-SSD不仅提升了SSD在车辆和行人检测下的准确度，更是提高了SSD的检测速度，从而验证了本文所使用的K-means+〖KG-*3〗+聚类算法的有效性和特征融合方法的高效性。

关键词: 单次检测器, 车辆和行人检测, K-means+〖KG-*3〗+聚类算法, 融合模块

Abstract: In view of the fact that the conventional improved SSD methods improve the object detection accuracy of SSD while reducing its detection speed, this paper proposes an improved KM-SSD method based on SSD. Firstly, K-means+〖KG-*3〗+ clustering algorithm is used to adaptively learn the ratio of width to height in prior boxes; Secondly, an efficient feature mergence module is designed to achieve high and low level feature information fusion; Finally, the KM-SSD method is verified on the challenging KITTI dataset. The experimental results show that the mAP of SSD is 62.7%, and its average detection time is 0.162 s; the mAP of KM-SSD is 69.8%, and its average detection time is 0.133 s. Consequently, KM-SSD method not only improves the accuracy of SSD in vehicle and pedestrian detection, but also improves the detection speed of SSD, which proves the effectiveness of K-means+〖KG-*3〗+ clustering algorithm and the efficiency of feature fusion method used in this paper.

Key words: SSD, vehicle and pedestrian detection, K-means+〖KG-*3〗+ clustering algorithm, mergence module

郑钦浩, 杨贞, 杨振 . 面向车辆和行人检测的KM-SSD方法[J]. 计算机与现代化, 2021, 0(03): 51-56.

ZHENG Qin-hao , YANG Zhen , YANG Zhen . KM-SSD Method for Vehicle and Pedestrian Detection[J]. Computer and Modernization, 2021, 0(03): 51-56.

参考文献

［1］王金强,黄航,郅朋,等. 自动驾驶发展与关键技术综述［J］. 电子技术应用, 2019,45（6）:28-36.
［2］罗会兰,陈鸿坤. 基于深度学习的目标检测研究综述［J］. 电子学报, 2020,48（6）:1230-1239.
［3］曹燕,李欢,王天宝. 基于深度学习的目标检测算法研究综述［J］. 计算机与现代化, 2020（5）:63-69.
［4］ LOWE D G. Distinctive image features from scale-invariant keypoints［J］. International Journal of Computer Vision, 2004,60（2）:91-110.
［5］ DALAL N, TRIGGS B. Histograms of oriented gradients for human detection［C］// Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. 2005:886-893.
［6］ VIOLA P, JONES M. Rapid object detection using a boosted cascade of simple features［C］// Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. 2001:511-518.
［7］ BAY H, ESS A, TUYTELAARS T, et al. Speeded-up robust features （SURF）［J］. Computer Vision and Image Understanding, 2008,110（3）:346-359.
［8］ KRIZHEVSKY A, SUTSKEVER I, HINTON G E. ImageNet classification with deep convolutional neural networks［J］. Communications of the ACM, 2017,60（6）:84-90.
［9］ DENG J, DONG W, SOCHER R, et al. ImageNet: A large-scale hierarchical image database［C］// 2009 IEEE Conference on Computer Vision and Pattern Recognition. 2009:248-255.
［10］GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation［C］// 2014 IEEE Conference on Computer Vision and Pattern Recognition. 2014:580-587.
［11］林菁,杨楠,臧勤. 基于FCN模型和选择性搜索的目标检测方法［J］. 雷达与对抗, 2019,39（2）:58-63.
［12］HE K M, ZHANG X Y, REN S Q, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015,37（9）:1904-1916.
［13］GIRSHICK R. Fast R-CNN［C］// 2015 IEEE International Conference on Computer Vision. 2015:1440-1448.
［14］REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with regionproposal networks［C］// International Conference on Neural Information Processing Systems. 2015:91-99.
［15］REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection［C］// 2016 IEEE International Conference on Computer Vision and Pattern Recognition. 2016:779-788.

［16］LIU W, ANGUELOV D, ERHAN D, et al. SSD: Single shot multibox detector［C］// European Conference on Computer Vision. 2016:21-37.

［17］〖ZK(#〗BI M X, QIAN Y M, YU K. Very deep convolutional neural networks for LVCSR［C］// The 16th Annual Conference of the International Speech Communication Association. 2015:3259-3263.
［18］周志华. 机器学习［M］. 北京:清华大学出版社, 2016.
［19］CAO G M, XIE X M, YANG W Z, et al. Feature-fused SSD: Fast detection for small objects［J］. Computer Vision and Pattern Recognition, 2018:arXiv,1709.05054.
［20］FU C Y, LIU W, RANGA A, et al. DSSD: Deconvolutional single shot detector［J］. Computer Vision and Pattern Recognition, 2017:arXiv,1701.06659.
［21］IOFFE S, SZEGEDY C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift［EB/OL］. （2015-03-02）［2020-07-16］. https://arxiv.org/abs/1502.03167.
［22］GEIGER A, LENZ P, STILLER C, et al. Vision meets robotics: The KITTI dataset［J］. International Journal of Robotics Research, 2013,32（11）:1231-1237.
［23］ EVERINGHAM M. The PASCAL visual object classes challenge［J］. Lecture Notes in Computer Science, 2005,111（1）:98-136.
［24］HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition［C］// 2016 IEEE Conference on Computer Vision and Pattern Recognition. 2016:770-778.
［25］HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks［C］// Proceedings of the 30th IEEE Conference on Computer Vision and Pattern Recognition. 2017:2261-2269.
［26］LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: Common objects in context［J］. Computer Vision, 2014,8693:740-755.