基于改进YOLOx的弱光照环境车辆检测方法  

Improved YOLOx-based vehicle detection method for low light environment

作　　者：杨晓寒 王峻 段中兴[1] 惠蕾蕾 YANG Xiaohan;WANG Jun;DUAN Zhongxing;HUI Leilei(College of Information and Control Engineering,Xi′an University of Architecture and Technology,Xi′an 710055,China;Traffic Engineering Construction Bureau of Jiangsu Province,Nanjing 210024,China;China Northwest Architecture Design and Research Institute Co.Ltd.,Xi′an 710018,China)

机构地区：[1]西安建筑科技大学信息与控制工程学院,陕西西安710055 [2]江苏省交通工程建设局,江苏南京210004 [3]中国建筑西北设计研究院有限公司,陕西西安710018

出　　处：《液晶与显示》2024年第6期801-812,共12页Chinese Journal of Liquid Crystals and Displays

基　　金：国家自然科学基金(No.51678470)。

摘　　要：在公路隧道等弱光照环境下,采集的车辆图像受外界因素影响较大,导致车辆检测精度低。针对此问题,提出了一种改进YOLOx算法的弱光照环境车辆实时检测方法。首先,基于引导滤波和区域能量特性融合准则对采集的车辆图像进行增强,解决图像中光照不均、目标轮廓信息模糊等问题。其次,基于Swin-Transformer网络结构,构建了改进YOLOx的车辆检测算法的主干网络,利用Transformer的全局建模能力对图像中的关键语义信息进行编码,强化网络细节特征的提取能力。同时,在颈部网络引入递归门控卷积替换空洞卷积,提高网络高层语义建模能力。最后,引入卷积注意力机制,加强网络对于低照度图像关键特征的提取与融合。在所构建的隧道车辆检测数据集UA-DETTUN上进行实验验证,结果表明,所提出方法的平均检测精度达到96.1%,相比于改进前的YOLOx算法提升了6.5%,同时网络的检测速度满足实时检测的要求,在车辆检测方面具有较高的应用价值。For the low light environment such as road tunnels,the acquired vehicle images are affected by external factors,which leads to low vehicle detection accuracy.For this problem,a real-time vehicle detection method for low light environment with improved YOLOx algorithm is proposed.Firstly,the collected vehicle images are enhanced based on the guiding filter and regional energy characteristic fusion criterion to solve the problems of uneven illumination and blurred target contour information in the images.Secondly,based on the Swin-Transformer network structure,the backbone network of the vehicle detection algorithm with improved YOLOx is constructed,and the global modeling capability of Transformer is used to encode the key semantic information in the images and strengthen the extraction capability of the network detail features.Meanwhile,recursive gated convolution is introduced to replace the null convolution in the neck network to improve the network’s high-level semantic modeling capability.Finally,a convolutional attention mechanism is introduced to enhance the network’s extraction and fusion of key features for low-illumination images.Experimental validation on the constructed tunnel vehicle detection dataset UA-DETTUN shows that the proposed method achieves an average detection accuracy of 96.1%,which is 6.5%better than the YOLOx algorithm before improvement,while the detection speed of the network meets the requirement of real-time detection.The proposed method has high application value in vehicle detection.

关 键 词：弱光照环境 Swin-Transformer 车辆检测 图像增强 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]