基于深度学习的无人驾驶汽车道路坑洞检测技术  

Road Pothole Detection for Autonomous Vehicles Based on Deep Learning

作　　者：周家强 胡宇航 石竞奇 宋森楠 阮良军[1] 

机构地区：[1]宁波工程学院机械与汽车工程学院,浙江 宁波

出　　处：《计算机科学与应用》2024年第12期29-35,共7页Computer Science and Application

基　　金：浙江省大学生科技创新活动计划暨新苗人才计划项目(2024R428A005)。

摘　　要：道路坑洞威胁着汽车的驾驶安全,针对无人驾驶汽车,进行可靠的坑洞检测尤为重要。在所有的检测方法中,基于深度学习网络的图像识别算法具有更高的精度和更快的检测速度。因此,本文旨在提供基于YOLOv5的高精度坑洞检测方法。具体来说,首先在中国城乡道路上采集了1000张图像,通过网络搜索引擎获得了1500张图片。使用YOLOv5s模型对获得的数据集进行训练,同时,对原有的模型进行了优化,增加SE注意力机制和BiFPN特征融合机制,以获得更好的精度和泛化性。检测结果表明,优化后的模型检测精度由81.1%提高到95.0%;mAP0.5由89.1%提高到92.2%;mAP0.5:0.95由48.2%提高到49.5%;检测速度基本与原模型持平,可满足实时检测要求。因此,本文获得了一种可实时且性能更优的道路坑洞检测方法,可应用于无人驾驶汽车安全系统。Road potholes pose a threat to the driving safety of cars, especially for autonomous vehicles, and reliable pothole detection is particularly important. Among all detection methods, image recognition based on deep learning networks has higher accuracy and faster detection speed. Therefore, this article aims to provide a high-precision pit detection method based on YOLOv5. Specifically, 1000 images were collected on urban and rural roads in China, and 1500 images were obtained through online search engines. Train the obtained dataset using the YOLOv5s model. Optimize the activation function, add attention mechanism and BiFPN feature fusion mechanism to achieve better accuracy and generalization. The results show that the accuracy of the optimized model has increased from 81.1% to 95.0%;mAP0.5 increased from 89.1% to 92.2%;mAP0.5:0.95 increased from 48.2% to 49.5%. The detection speed is basically the same as the original model, which can meet the requirements of real-time detection. Therefore, this article presents a real-time and high-performance method for detection, which can be applied to the safety sy

关 键 词：深度学习 YOLOv5 注意力机制 特征融合机制 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]