立体视觉的目的之一就是为了获得周围场景的3维信息,其关键在于匹配算法。然而即便是使用目前先进的通用处理器,其计算致密视差图所需的时间仍无法满足高速自主导航的需求。为了解决这个问题,提出了一种基于现场可编程门阵列(FPGA)的双目立体视觉系统的设计方案,同时介绍了系统的硬件结构,并在讨论区域匹配的快速算法的基础上,提出了基于FPGA的像素序列和并行窗口算法框架,用以实现零均值像素灰度差平方和(ZSSD)的匹配算法。该算法是先将视频信号经解码芯片生成场景立体图像对,并由FPGA来完成立体图像对的几何校正和ZSSD匹配算法,然后将获得的致密视差图通过PC I总线发送至上位机。实践表明,该算法效果好、速度快,不仅具有较强的鲁棒性,并且硬件系统性能稳定、可靠。此外,该方案还适用于像素灰度差的绝对值和(SAD)和像素灰度差的平方和(SSD)等多种传统区域匹配算法的快速实现和实时处理。

The goal of stereo vision is to recover the three-dimensional information of a scene,and the core of stereo vision is to find the corresponding pixels.However,the correspondence search is too time-consuming for high speed robot vehicles autonomous navigation,even if state-of-art general-purpose processors are used to accelerate it.Aiming at this problem,this paper presents a design scheme for binocular stereo vision system based on field programmable gate arrays(FPGA),its hardware structure is introduced,and a fast zero mean sum of squared differences(ZSSD) stereo matching algorithm is discussed.Moreover,a pixel-serial and window-parallel architecture based on FPGA processor is proposed to achieve ZSSD matching algorithm,which is suitable for parallel processing.The stereopsis is captured by two same decoder chips.After rectification and ZSSD matching implemented by FPGA,dense disparity map is computed and sent to general-purpose computer by PCI bus.The proposed scheme is very robust and exhibits great performance,such as high speed.And the hardware system has high stability and reliability.In addition,this scheme is also applicable to fast and real-time processing of other conventional area-based stereo matching algorithms,such as sum of absolute differences(SAD),sum of squared differences(SSD),etc.