基于高速前向运动视频的运行环境检测模型

项目名称： 基于高速前向运动视频的运行环境检测模型

项目编号： No.61272354

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 自动化技术、计算机技术

项目作者： 罗四维

作者单位： 北京交通大学

项目金额： 80万元

中文摘要： 为保证高速交通运载工具的安全运行，利用计算机视觉感知技术快速检测环境状态已经成为一种重要的手段。本项目拟以高速铁路运行沿线环境为背景，建立基于高速前向运动视频的环境检测模型。项目涉及图像恢复、增强等经典问题以及高速前向运动视频分割、全景图拼接等崭新课题；主要研究内容：(１)提出一种新的基于高速前向运动视频的全景图拼接技术，达到以全景图替代视频图像实现场景图像信息压缩和通过全景图比对来发现异常的目的；（2）前向运动视频图像的快速分割以及帧图像快速展开是实现全景图快速并行拼接与比对的前提，拟以光流技术为基础，提出新的融合先验的光流法；（3）研究基于流形的正则化方法和偏微分方程模型，实现适用于高速运行环境下图像的恢复、增强。该研究可为建立高铁沿线环境的安全预警系统提供技术理论支持，且该模型可以推广到公路交通、隧道等其它环境，本项目的全景图拼接技术可为三维全景虚拟现实开辟一条新的制作途径。

中文关键词： 计算机视觉；前向运动视频；全景图拼接；异常检测；

英文摘要： Recently, machine vision-based high-speed railway maintenance and surveillance has attracted many interests to ensure the safety of high-speed railway transportation. Therefore, this project aims at constructing a novel front view railway video-based maintenance system, and then focuses on solving scientific problems in the field of image restoration, image enhancement, video segmentation and image panorama. The main research topics include: (1) proposing a new image panorama approach based on single high-speed front view video to generate the non-overlapped scenes. It can largely reduce the data size and further can be used to detect the abnormal objects; (2) developing a new optical flow method with railway priors to deal with video segmentation and image unwarping that is the fundamental task for image panorama; (3) presenting an image restoration and enhancement model, which combines the manifold-based regularization and partial difference equation, to highly improve the image quality. The theoretical results of this project can be used to construct the railway surveillance system to guarantee the safety of high-speed railway environment. Then, such model can be easily generalized to other transportations, such as high way and tunnel. Moreover, the proposed image panorama technology would provide a new idea

英文关键词： computer vision；forward looking video；panoramic stitching；abnormality detection；