基于改进SSD算法的交通标识检测方法研究  

Research on traffic sign recognition method based on improved SSD algorithm

作　　者：詹华伟 邹昊好[1,2] 刘旭 史水娥 Zhan Huawei;Zou Haohao;Liu Xu;Shi Shui'e(College of Electronic and Electrical Engineering,Henan Normal University,Xinxiang 453007,China;Henan KeyLaboratory of Optoelectronic Sensing Integrated Application,Henan Normal University,Xinxiang 453007,China)

机构地区：[1]河南师范大学电子与电气工程学院,新乡453007 [2]河南师范大学河南省光电传感集成应用重点实验室,新乡453007

出　　处：《电子测量技术》2022年第17期79-85,共7页Electronic Measurement Technology

基　　金：河南省高等学校重点科研项目基础研究项目(19B510006)资助。

摘　　要：针对目前SSD算法对小目标检测精确度低,泛化能力弱,且存在误检、漏检等问题,提出一种基于SSD网络的交通标识检测方法。为增加对目标的检测精度,使用ResNet-50网络作为SSD算法的骨干网络,在额外添加层中加入BN层,提高训练速度;使用sub-pixel来代替上采样,提高识别目标分辨率,并加入MFPN模型融合低层与高层特征信息,避免出现漏检问题。实验结果表明与现有的SSD算法相比,改进的SSD算法在公开数据集CCTSDB和GTSDB数据集上mAP值分别提高4.2%和3.1%,FPS保持在87.2 f/s,检测精度显著提升。满足对交通标识实时检测的要求,在无人驾驶领域具有广泛的应用前景。This paper proposes a traffic sign detection method based on an SSD network.This method improves the existing SSD algorithm,which has low detection accuracy and weak generalization ability for small targets,and has problems such as false detection and missed detection.The ResNet-50 network is used as the backbone network of the SSD algorithm,and the BN layer is added to the additional layer to improve the training speed.Sub-pixel is used instead of upsampling to improve the resolution of the recognition target,and the MFPN model is added to fuse the low-level and high-level feature information to avoid the problem of missed detection.The experimental results show that the improved SSD algorithm improves the mAP value by 4.2%and 3.1%on the public datasets CCTSDB and GTSDB datasets,respectively,the FPS remains at 87.2 f/s,and the detection accuracy is significantly improved.This work meets the requirements for real-time detection of traffic signs and has broad application prospects in the field of unmanned driving.

关 键 词：SSD 卷积网络 MFPN模型 SUB-PIXEL CCTSDB 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]