大工至善|大学至真分享 http://blog.sciencenet.cn/u/lcj2212916

博文

[转载]【信息技术】用于道路环境中物体场景理解的分类系统在汽车领域的应用

已有 1633 次阅读 2019-9-1 11:26 |系统分类:科研笔记|文章来源:转载


本文为美国普林斯顿大学(作者:John Valentino)的硕士论文,共96页。

 

本文探讨了利用计算机视觉中的方法来开发一个简单但可扩展且有效的分割、跟踪和分类系统,以便标记和理解道路环境中的场景和对象。该项目利用了安装在运动型多用途车车顶上的彩色单目摄像头,旨在扩展到具有立体声功能的平台,如PAVE Prospect 12车辆。首先从一个静止的位置出发,利用背景模型进行跟踪、检测,描述并尝试了一种基于运动的方法,该方法基于物体相对于车辆的推测速度来进行分离。探索了一种结合道路结构、利用概率模型和标号训练快速分析场景的新方法,并将该系统集成起来,为选课的测试视频文件提供基本的跟踪和分类。最后,该项目提出了如何通过Prospect 12车辆平台(例如车辆速度、GPS、立体摄像头、互联网接入)实现访问功能的建议,以克服当前的局限性,促进大幅度提高潜在的性能。

 

This paper explores the use of methods incomputer vision to develop a simple but extensible and effective segmentation,tracking, and classification system for the purpose of labeling andunderstanding scenes and objects in the road environment. The project makes useof a color monocular camera mounted to the roof of a sports utility vehicle andis intended to be extendable to platforms with stereo capabilities such as thePAVE Prospect 12 vehicle. Tracking and detection are first explored from astationary position that makes use of a background model and a motion basedapproach that separates objects based on their inferred velocity relative tothe vehicle is described and attempted. A novel method of incorporating roadstructure and quickly analyzing the scene using probabilistic models andlabeled training scenes is explored, and the system is integrated to providebasic tracking and classification on testing video files for selected classes.The project concludes with suggestions on how functionality accessible via theProspect 12 vehicle platform (e.g. vehicle velocity, GPS, stereo cameras,internet access) could be incorporated to substantially increase performanceand build upon weaknesses of this implementation.

 

引言

物体分割

车辆运动

粗糙场景分类器

个体对象分类

集成与跟踪

性能改进与提升

附录图像列表

附录程序清单


更多精彩文章请关注公众号:qrcode_for_gh_60b944f6c215_258.jpg



https://blog.sciencenet.cn/blog-69686-1196146.html

上一篇:[转载]【计算机科学】【2017.09】【含源码】基于深度学习的农业领域实例分割
下一篇:[转载]【计算机科学】【2017.07】事件驱动数据的深度神经网络与硬件系统
收藏 IP: 220.178.172.*| 热度|

0

该博文允许注册用户评论 请点击登录 评论 (0 个评论)

数据加载中...
扫一扫,分享此博文

Archiver|手机版|科学网 ( 京ICP备07017567号-12 )

GMT+8, 2024-9-19 22:18

Powered by ScienceNet.cn

Copyright © 2007- 中国科学报社

返回顶部