CASIA OpenIR  > 毕业生  > 硕士学位论文
汽车辅助驾驶系统中的单目视觉导航关键技术研究
黄冠
学位类型工程硕士
导师王欣刚
2016-05
学位授予单位中国科学院大学
学位授予地点北京
关键词汽车辅助驾驶系统、车道线检测、车辆检测、车辆跟踪、卷积神经网络
摘要近年来,随着人工智能、计算机视觉以及深度学习的发展,汽车辅助驾驶系统已成为十分火热的研究和应用领域。在汽车辅助驾驶系统中,车道保持、偏航预警、自适应巡航、前车距离测量、碰撞预警等关键功能都依赖于基于单目的计算机视觉技术。因此本论文的研究内容具有重要的理论意义和应用价值。
基于以上背景,本文针对汽车辅助驾驶系统中的单目视觉关键技术进行了深入的探讨和研究,实现了车道线检测、路面天空区域识别、车辆检测、车辆跟踪等关键模块,并且进行了深入的分析和创新。同时,对于前方车辆定位问题,结合上面的单目视觉关键模块,本文提出了一套完整的整体框架,实现了实时鲁棒准确的前方车辆定位与跟踪。
具体的,本文完成的工作和取得的研究成果包括以下几个方面:
(1)提出了一种对前车定位的目标区域(ROI)进行缩减的方法。具体包括车道线检测与跟踪、路面区域识别与天空区域识别。结合车道线信息、路面区域信息以及天空区域信息,对前方车辆在图像中的位置实现了约束,缩减了感兴趣的区域,对提高前车定位的准确性和实时性具有非常重要的作用。
(2)研究并实现了实时鲁棒准确的车辆检测模块。本文结合传统的AdaBoost级联分类器以及类Haar特征的目标检测方法和基于卷积神经网络的目标检测方法,设计并实现了一种快速准确鲁棒的车辆检测算法。
(3)研究并实现了快速准确的前方车辆跟踪模块。本文结合卷积神经网络的方法以及基于通过检测来跟踪(tracking by detection)的相关滤波器的方法,设计并实现了一种非常快速、准确的前方车辆跟踪算法。
(4)对于汽车辅助驾驶中的前车定位与跟踪问题,结合车道线检测、路面天空区域识别、车辆检测以及车辆跟踪等模块,搭建了一套实时鲁棒的系统。
本文实现了高级汽车辅助驾驶系统(ADAS)中基于单目视觉的关键模块与技术,同时满足系统的精度和实时性的要求,对于汽车辅助驾驶的研究和实际应用具有一定的参考和借鉴意义。
其他摘要In recent years, with the development of artificial intelligence, computer vision and deep learning, advanced driver assistant system has become a very popular research and application field. The key functions of advanced driver assistant system, including lane keeping, yaw warning, adaptive cruise control, distance measurement of leading vehicle and collision warning, all depend on monocular computer vision technologies. So the research contents of this paper have important theoretical significance and application values.
Based on the above background, the key technologies of the monocular vision in advanced driver assistant system are discussed and studied in this paper. We implement and innovate some key modules including lane detection and tracking, road and sky region recognition, vehicle detection and vehicle tracking. At the same time, based on key modules introduced above, we present a framework for leading vehicle localization and tracking, which is real-time and accurate.
Specifically, the work done and the research results obtained in this paper include the following aspects:
(1) A method to reduce the region of interest for the problem of leading vehicle localization and tracking is proposed. The method for ROI region reduction includes lane detection and tracking, road region recognition and sky region recognition.
(2) A real-time and accurate vehicle detection module is introduced in this paper. Based on the conventional object detection method of Haar-like features and cascaded adaboost classifier, and convolutional neural network, we design a real-time and accurate vehicle detection system.
(3) A fast and accurate vehicle tracking module is proposed. Based on convolutional neural network trained for vehicle detection and adaptive correlation filters for visual tracking, we propose a fast and effective vehicle tracking algorithm.
(4) A real-time and accurate system is built for the leading vehicle localization and tracking problem, combined with lane detection, road and sky recognition, vehicle detection, vehicle tracking and other modules.
In this paper, the key modules and technologies based on monocular vision in the advanced driver assistant system (ADAS) are realized. At the same time, the requirements of the system's precision and real-time performance are also met. The contents of this paper are supposed to give some consult and reference meaning for the research and practical application of the advanced driver assistant system.
语种中文
文献类型学位论文
条目标识符http://ir.ia.ac.cn/handle/173211/11789
专题毕业生_硕士学位论文
作者单位中国科学院自动化研究所
推荐引用方式
GB/T 7714
黄冠. 汽车辅助驾驶系统中的单目视觉导航关键技术研究[D]. 北京. 中国科学院大学,2016.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
汽车辅助驾驶系统中的单目视觉导航关键技术(2680KB)学位论文 暂不开放CC BY-NC-SA请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[黄冠]的文章
百度学术
百度学术中相似的文章
[黄冠]的文章
必应学术
必应学术中相似的文章
[黄冠]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。