基于三维差时投影法的超分辨率车辆重建算法研究

项目名称： 基于三维差时投影法的超分辨率车辆重建算法研究

项目编号： No.61004110

项目类型： 青年科学基金项目

立项/批准年度： 2011

项目学科： 交通运输

项目作者： 张伟

作者单位： 华中科技大学

项目金额： 7万元

中文摘要： 本课题拟将三维车辆模型与超分辨率重建技术相结合，深入研究利用单个监控相机重建车辆的特征提取算法以及三维重建问题,重点分析三维车辆模型与二维车辆图像之间的投影模型, 解决超分辨率技术中的成像模型估计和像素配准问题，突破了现有的仅局限于二维平面的图像配准模型,从而为车辆监控系统中的超分辨率图像重建算法提供新的思路。成像模型本质上应反映三维真实世界到二维图像的投影过程，而本课题将三维车辆模型引入来估计成像模型与像素配准，可以实现高精度的图像配准。本课题在研究交通监控系统的基础上，首先根据车辆目标的多帧图像，重建其初始三维模型，而车辆的精确三维模型将通过一个分类算法从三维车辆模型库中取得。得到车辆三维模型后，我们依据一个三维差时投影法取得图像的成像模型以及像素配准，并使用经典的迭代反向投影法重建高分辨率图像。本课题将在二维图像特征提取，三维目标分类、图像配准与成像模型估计等方面做出创新性工作。

中文关键词： 交通监控；超分辨率图像重建；三维重建；图像匹配

英文摘要： This Project will combine the 3D vehicle model and the super-resolution image reconstruction technique, and focus on the vehicle feature extraction and 3D vehicle reconstruction. The projection between the 3D vehicle model and 2D vehicle image will be analyzed. The problem of imaging model estimation and pixel registration will also be solved and a breakthrough is made in the 2D image plane-based image registration. Thus we provides a new method for the super-resolution reconstruction in the vehicle surveillance systems. Imaging model is defined as the projection from 3D real-world scenes to 2D image plane, and the 3D vehicle model is included in this project to obtain high-precision pixel registration. This project is based on our research on the traffic surveillance system. Firstly, we reconstruct the vehicle's initial 3D model according to its video frames, and vehicle's accurate 3D model is obtained from a 3D vehicle model database by using a classification algorithm. Secondly, we employ a time-difference project algorithm to obtain the imaging model and pixel registration, and the classical iterative back projection method is utilized to reconstruct the high-resolution image. In this project, the original work will be made in the 2D image feature extraction, 3D object classification, pixel registration, and imaging model estimation.

英文关键词： Traffic surveillance; Super-resolution