基于机器视觉的雨雾天驾驶辅助系统设计  

Design of Driving Assistance System under Rain and Haze Weather Based on Machine Vision

作　　者：洪儒 于力涵 戴安邦 谢迎娟[1] HONG Ru;YU Lihan;DAI Anbang;Xie Yingjuan(School of Information Science and Engineering,Hohai University,Changzhou 213002,China)

机构地区：[1]河海大学信息科学与工程学院,江苏常州213002

出　　处：《计算机测量与控制》2024年第3期259-266,共8页Computer Measurement &Control

基　　金：国家自然科学基金(61573128);国家自然科学基金(61701169);国家重点研发计划(2018YFC0407101);教育部产学合作协同育人项目(220803494162012,220603632072407);国家级大学生创业训练项目(202210294232E);江苏省大学生创新创业训练计划项目(202210294220Y);河海大学本科实践教学改革研究项目(河海教务[2022]47号);河海大学创新性实验项目(河海教务[2022]49号)。

摘　　要：随着私家车数量日益增多,雨雾天交通安全问题成为了亟待解决的难题;在嵌入式硬件资源有限的情况下,为驾驶用户设计了基于机器视觉的雨雾天语音辅助驾驶系统;系统结合了湿度传感器、轻量化去雾神经网络AOD-NET和目标检测模型YOLOv5n;在目标检测模型YOLOv5n上,利用K-means++算法重新设计锚框,选取较优的骨干网络并利用模型剪枝进一步压缩模型大小;实验结果表明,改进的模型在Jetsonnano上的FPS达到了17.78,最终mAP在人工加雾、分辨率变化的TT100K (Tsinghua-Tencent 100K)数据集到达了65.8%,满足了正常天气与雨雾天气下的驾驶辅助实际应用。With the increase of private cars,traffic safety in rainy and foggy days has become an urgent problem to be solved.A voice assisted driving system for driving users in rainy and foggy weather based on machine vision is designed,with the limited embedded hardware resources.The system combines the humidity sensors,lightweight dehazing neural network AOD-NET,and object detection model YOLOv5n.The K-means++algorithm is used to redesign the anchor frame on the object detection model YOLOv5n.The optimal backbone network is selected to further compress the model size by using the model pruning.The experimental results show that the FPS of the improved model on the Jetson nano reaches by 17.78,and the final mAP by 65.8%on the artificially fogged and resolution changed TT100K(Tsinghua-Tencent 100K)dataset,meeting the practical application of driving assistance in normal weather and rainy and foggy weather.

关 键 词：机器视觉 Jetson nano 雨雾天辅助驾驶 交通标志检测 AOD-NET去雾 YOLOv5n 

分 类 号：V19[航空宇航科学与技术—人机与环境工程]