基于立体R-CNN的自动驾驶中的路障识别与定位
Roadblock Recognition and Location in Autonomous Driving Based on Stereo R-CNN
刘启 1卢林 2朱兆旻3
作者信息
- 1. 北部湾大学机械与船舶学院 广西钦州市 535011
- 2. 广西大学 计算机与电子信息学院 广西南宁市 530004
- 3. 桂林航天工业学院电子信息与自动化学院 广西桂林市 541004
- 折叠
摘要
基于深度学习的自动驾驶技术是目前最为先进和主流的自动驾驶技术之一,它通过应用深度学习算法来实现对驾驶场景的感知、路径规划、行为决策和运动控制,极大地提高了自动驾驶汽车的安全性和可靠性.文章提出了一种新型的立体区域卷积神经网络模型对自动驾驶中的路障的识别与定位进行研究.结果显示,文章提出的方法在图像识别精度和速度上要优于Mono3D和VeloFCN等其它模型.
Abstract
Deep-learning based autonomous driving technology is currently one of the most advanced and mainstream autonomous driving technologies.It realizes the perception of driving scene,path planning,behavior decision-making and motion control by applying deep learning algorithm,which greatly improves the safety and reliability of autonomous vehicle.This paper focuses on the problem of obstacle recognition and localization in autonomous driving,and uses a three-dimensional region convolutional neural network to study the recognition and localization of obstacles in autonomous driving.The results show that the Stereo R-CNN method outperforms other Mono3D and VeloFCN models both in image recognition accuracy and velocity.
关键词
深度学习/R-CNN/自动驾驶/路障识别与定位Key words
Deep learning/R-CNN/Autonomous driving/Roadblock Identification and Location引用本文复制引用
基金项目
广西自然科学基金(2021GXNSFAA220048)
出版年
2024
NETL
NSTL
万方数据