【计算机科学】【2019】基于深度学习的车辆相关场景理解

本文为新西兰奥克兰理工大学（作者：Xiaoxu Liu）的硕士论文，共110页。

自动驾驶技术是未来交通发展的必然趋势，也是人工智能领域的杰出成就之一。主要是深度学习对自动驾驶的发展做出了重大贡献。深度学习不仅能促进自主车辆感知、识别周围环境，还能识别、分类与车辆相关的各种信息。

随着深度学习技术的不断升级和完善，它能够被及时、便捷地学习和运用。大量的预训练网络和公共数据集为训练大量的交通场景提供了方便。然而，根据各国的交通规则和交通设施，自动驾驶技术还不足以灵活地理解复杂交通环境中的场景。目前还没有针对所有交通场景设计的算法。在这篇论文中，我们主要处理的问题是了解在车辆相关的场景中如何使用深度学习。据我们所知，这是第一次利用奥克兰交通环境作为场景理解的分析对象。此外，将自动场景分割与目标检测相结合，实现了交通场景的理解。基于深度学习的技术大大减少了人类的操作。此外，本项目中的数据集提供了大量奥克兰交通数据。同时，结合车辆检测结果巩固了CNN处理的性能。

Automated driving technology is aninevitable trend in the future development of transportation, it is also one ofthe eminent achievements in the matter of artificial intelligence. Primarilydeep learning produces a significant contribution to the progression ofautomatic driving. Deep learning not only promotes autonomous vehicles to senseand identify the surrounding environment, but also identifies and classifiesvarious information regarding to vehicles. With the upgrades and improvement ofdeep learning technology, it can be promptly and readily learned and employed.A large number of pretraining networks and public datasets have providedconvenience for training numerous traffic scenes. Nevertheless, automateddriving technology is not flexible enough to understand scenes in complextraffic environments, with regard to traffic rules and transportationfacilities in various countries. There is no algorithm so far designed for alltraffic scenes. In this thesis, our contributions are that we primarily dealwith the issue of understanding of vehicle-related scene using deep learning.To the best of our knowledge, this is the first time that we utilize Aucklandtraffic environment as an analysis object for scene understanding. Moreover,automatic scene segmentation and object detection are coalesced for trafficscene understanding. The techniques based on deep learning dramaticallydecrease human manipulations. Furthermore, the datasets in this project providea large amount of Auckland traffic data. Meanwhile, the performance of CNNprocessing is consolidated by combining with vehicle detection outcome.

1. 引言
2. 文献回顾
3. 研究方法
4. 研究结果
5. 分析与讨论
6. 结论与展望

更多精彩文章请关注公众号：