| 613 | 7 | 480 |
| 下载次数 | 被引频次 | 阅读次数 |
交通信号灯的准确检测与识别可以提高驾驶的安全性,减少交通事故的发生。为了提高移动端识别的准确率和速度,提出一种改进的轻量级YOLOv3模型实现交通信号灯的检测与识别。首先,采用轻量级的ShuffleNetv2网络替换YOLOv3的主干网络DarkNet53,实现交通信号灯的快速检测与识别;接着,融合ShuffleNetv2网络中的低、中、高层特征组成最终的高层输出特征,以丰富交通信号灯的特征表示;最后,基于多尺度的检测和结果融合,完成交通信号灯的准确检测与识别。实验证明,本文提出的轻量级YOLOv3模型应用于交通信号灯公开数据集LaRa时,平均精度均值达到87.21%,漏检率仅为6.24%。与其他轻量级YOLOv3模型相比,本文的模型也有着更高的平均精度均值和更低的漏检率,并且模型大小仅为YOLOv3的1/8,检测速度却是YOLOv3的3倍。
Abstract：The detection and recognition technology of traffic lights can improve the safety of driverless cars and reduce the occurrence of traffic accidents. In order to improve the accuracy and speed of traffic light detection and recognition in the vehicle, an improved lightweight YOLOv3 model for traffic lights detection and recognition was proposed. Firstly, the YOLOv3 backbone DarkNet53 was replaced by the ShuffleNetv2 of lightweight to achieve the accurate detection and recognition of traffic lights. Secondly, the features from low, middle and high levels in ShuffleNetv2 network were fused as the final high-level feature output to enrich the traffic light features in ShuffleNetv2.Finally, based on the multi-scale detection and result fusion the proposed model finished the traffic light detection and recognition. The results show that the lightweight YOLOv3 model achieves 87.21% mean average precision and 6.24%missing detection rate on LaRadataset. Compared with other lightweight YOLOv3 models, the proposed YOLOv3 model has higher mean average precision and lower missing detection rate. Besides, the size of the model is reduced to one-eighth of YOLOv3, and the speed is three times faster than that of YOLOv3.
[1] OMACHI M, OMACHI S. De tection of traffic light using structural information[C]//Proceedings of the IEEE 10th International Conference on Signal Processing, October 24-28, 2010, Beijing, China. New York:IEEE Xplore, 2010:809-812.
[2]许明文，张重阳.基于显著性特征的交通信号灯检测和识别[J].计算机与数字工程，2017, 45(7):1397-1401.
[3] KIM Y K, KIM K W, YANG X L. Real time traffic light recognition system for color vision deficiencies[C]//Proceedings of the International Conference on Mechatronics and Automation, August 5-8, 2007, Harbin, China. New York:IEEE Xplore, 2007:76-81.
[4] KIM J, CHO H, HWANGBO M, et al. Deep traffic light detection for self-driving cars from a large-scale dataset[C]//Proceedings of the 2018 21st International Conference on Intelligent Transportation Systems(ITSC), November 4-7,2018, Maui, HI, USA. New York:IEEE Xplore, 2018:280-285.
[5] M譈LLER J, DIETMAYER K. Detecting traffic lights by single shot detection[C]//Proceedings of the 2018 21st International Conference on Intelligent Transportation Systems(ITSC), November 4-7, 2018, Maui, HI, USA. New York:IEEE Xplore, 2018:266-273.
[6] LIU W, ANGUELOV D, ERHAN D, et al. SSD:single shot multibox detector[C]//Proceedings of the 14th European Conference on Computer Vision, October 11-14, 2016,Amsterdam, The Netherlands. Berlin:Springer, 2016:21-37.
[7] HAN C, GAO G Y, ZHANG Y. Real-time small traffic sign detection with revised faster-RCNN[J]. Multimedia Tools and Applications, 2019, 78(10):13263-13278.
[8] REN S Q, HE K M, GIRSHICK R, et al. Faster R-CNN:towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and 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]//Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), June 27-30, 2016, Las Vegas, NV, USA. New York:IEEE Xplore, 2016:779-788.
[10]王思雨.基于YOLO的交通目标检测方法[D].北京：华北电力大学，2019.
[11] REDMON J, FARHADI A. YOLO9000:better, faster, stronger[C]//Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), July 21-26, 2017, Honolulu, HI, USA. New York:IEEE Xplore,2017:6517-6525.
[12] REDMON J, FARHADI A. YOLOv3:an incremental improvement[EB/OL].(2018-12-13)[2021-03-13]. https：//arxiv.org/abs/1804.02767.
[13]董晓玉，孔斌，杨静，等.小尺度交通信号灯的检测与状态识别[J].测控技术，2020, 39(11):45-51.
[14]王莹，丁鹏.基于深度学习的交通信号灯检测及分类方法[J].汽车实用技术，2018(17):89-91.
[15] MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2:practical guidelines for efficient CNN architecture design[C]//Proceedings of the Europesn Conference on Computer Vision(ECCV), Munich, Germany. Berlin:Springer,2018:116-131.
[16] IOFFE S, SZEGEDY C. Batch normalization:accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on Machine Learning(ICML), July 11-16, 2015, Lille, France.Berlin:Springer, 2015:448-456.
[17] ISAAC B J, KINJO H, NAKAZONO K, et al. Training of multi-layered neural network for data enlargement processing using an activity function[J]. Journal of Electrical Engineering, 2019(1):1-7.
[18]于祥.深度学习与飞桨[M].北京：人民邮电出版社，2019:179.
[19]钱弘毅，王丽华，牟宏磊.基于深度学习的交通信号灯快速检测与识别[J].计算机科学，2019, 46(12):272-278.
[20]孙颖，张毅，丁卫平，等.基于近似骨架和混合蛙跳算法的K-means方法及其在眼底病历图像中的应用[J].南通大学学报（自然科学版），2020, 19(2):36-42.
[21] de CHARETTE R, NASHASHIBI F. Traffic light recognition using image processing compared to learning processes[C]//Proceedings of IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS), October 11-15,2009, Louis, MO, USA. New York:IEEE Xplore, 2009:333-338.
基本信息:
DOI：10.12194/j.ntu.20201013001
中图分类号:U463.6;TP391.41
引用信息:
[1]邵叶秦,周昆阳,郑泽斌,等.基于改进的轻量级YOLOv3的交通信号灯检测与识别[J],2021,20(03):34-40.DOI:10.12194/j.ntu.20201013001.
基金信息:
国家自然科学基金面上项目（61671255）;; 国家级大学生创新训练计划项目（202110304050Z,202110304047Z）;; 江苏省大学生创新训练计划项目（201910304158H,202010304180H,202010304122Y）;; 南通市科技计划项目（MS12020078）
2021-03-22
2021
2021-04-27
2021
1