基于深度学习的夜间危险驾驶行为检测算法  

Detection Algorithm of Dangerous Driving Behavior at Night Based on Deep Learning

作　　者：唐天俊 宋平 TANG Tian-jun;SONG Ping(Chongqing Technology and Business Institute,Chongqing 401520,China;Sichuan Xingshu Engineering Survey and Design Group Co.,Ltd.,Chengdu 610000,China)

机构地区：[1]重庆工商职业学院,重庆401520 [2]四川兴蜀工程勘察设计集团有限公司,四川成都610000

出　　处：《电脑与信息技术》2024年第4期9-13,共5页Computer and Information Technology

摘　　要：夜间弱光环境下的危险驾驶行为易导致交通事故的发生,然而,目前多数危险驾驶行为检测算法研究主要集中于光照充足的环境,而在弱光环境下的检测准确率偏低。针对该难点,提出了一种基于深度学习的夜间危险驾驶行为检测算法。该算法由弱光增强模块和检测模块构成。其中,弱光增强模块采用轻量化的零参考深度曲线估计算法提高图像曝光度,检测模块基于Nano Det-Plus模型检测弱光增强处理后的图像是否存在危险驾驶行为。实验结果表明,所提算法在夜间弱光环境下具有较高的检测准确率,同时模型参数量小,检测速度可达毫秒级,可部署在移动设备上进行实时检测。Dangerous driving behavior in low light environment at night is easy to lead to traffic accidents.However,most of the current research on dangerous driving behavior detection algorithms mainly focuses on the environment with sufficient light,and the detection accuracy in low light environment is low.To solve this problem,a deep learning-based night dangerous driving behavior detection algorithm is proposed.The algorithm consists of two modules:the weak light enhancement module and the detection module.Among them,the low-light enhancement module uses a lightweight zero-reference depth curve estimation algorithm to improve image exposure,and the detection module detects whether there is dangerous driving behavior in the images after low-light enhancement based on the NanoDet-Plus model.The experimental results show that the proposed algorithm has high detection accuracy in low light environment at night,and the number of model parameters is small,the detection speed can reach the millisecond level,and it can be deployed on mobile devices for real-time detection.

关 键 词：深度学习 弱光环境 夜间危险驾驶 行为检测 弱光增强 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]