双目立体相机中实时匹配算法的FPGA实现  

Real- time matching algorithm in binocular stereo camera and its implementation with FPGA

作　　者：单洁[1] 唐垚[1] 姜晖[1] 

机构地区：[1]西安邮电大学通信与信息工程学院,陕西西安710121

出　　处：《激光与红外》2013年第12期1406-1410,共5页Laser & Infrared

基　　金："973";"863"项目;国家重大科技专项(No.61271292);陕西省科技厅工业攻关计划(No.2012K06-27);西安邮电大学中青年基金(No.109-0425;109-04367)资助

摘　　要：在双目立体相机中,利用图像处理计算场景深度信息是一项关键技术。通过研究立体视觉图像匹配原理,提出一种基于FPGA的立体图像实时匹配算法的实现方法。该算法以Census变换为基础借助于像素在邻域中灰度相对值的排序进行相似度比较,来实现区域立体匹配;在左右一致性约束下采用多窗口相关匹配方法改善深度不连续图像的匹配质量,提高匹配准确度。利用FPGA流水线和并行处理技术实现了双目立体相机的实时图像匹配。结果表明,该图像匹配结构具有较高的吞吐率和处理速度,可以工作在97.3MHz频率下实现1024×1024灰度图像30f/s的实时处理。To compute the depth information using image processing is one of key techniques in the binocular stereo camera. Stereo vision image matching principle is analyzed and a new method of realizing real-time stereo image matching algorithm based on FPGA is proposed to decrease the matching time. In order to achieve regional stereo matching,based on the Census transform,the similarity are compared through the gray level of the pixel in the neighborhood; Multiple window correlation matching method are used to improve the quality of image when depth does not match on the left- right consistency constraints,thus,the matching accuracy is improved. Real- time image matching is implemented by using FPGA pipelining and parallel processing technology. The experimental results demonstrate that the method has higher throughput rate and processing speed,the real- time processing of 1024 × 1024 gray image with 30fps are realized at the frequency of 97. 3MHz.

关 键 词：双目立体相机 实时图像匹配 多窗口相关匹配 FPGA 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN911.73[自动化与计算机技术—计算机科学与技术]