摘要
立体匹配是双目视觉系统分析中的重要环节之一,直接决定三维信息重建的质量和效率.为提升立体匹配实时性与准确性,提出一种基于图像先验信息的立体匹配算法.算法首先采用BP神经网络MIV方法提取并筛选图像有效特征值,并以视觉系统应用环境不同将图像数据分为的简单背景图像和复杂背景图像,然后在测距 1-2m的实验室条件,利用双目摄像头和CORE I7 处理器采集图像数据,并在Visual Studio 2015 中按照3:2 对图像进行大小裁剪,最后基于BM优化算法与SGBM改进算法对图像进行立体匹配.简单背景仿真结果表明,未裁剪BM优化算法的测量误差未 0.9%,仿真时间为2s,较其它算法而言,仿真时间最短,实时性最高;复杂背景仿真显示,裁剪后的SGBM改进算法,测量误差为 0.4%,仿真时间2.5s,测量误差大幅降低.在图像先验信息的基础上,通过优化BM算法提高了立体匹配实时性,基于改进SGBM算法提高了立体匹配准确性,为双目视觉系统的实际应用提供了理论依据.
Abstract
Stereo matching is one of the important links in the analysis of binocular vision system,which directly determines the quality and efficiency of three-dimensional information reconstruction.In order to improve the real-time and accuracy of stereo matching,this paper proposes a stereo matching algorithm based on image prior informa-tion.Firstly,the algorithm uses BP neural network MIV method to extract and filter the effective feature values of the image,and divides the image data into simple background images and complex background images according to the different application environments of the vision system.And then=It usesbinocular cameras and COREI7 processor to collect image data under the laboratory conditions of 1-2m ranging.In Visual Studio 2015,the image is cropped ac-cording to the size of 3:2.Finally,the image is matched based on BM optimization algorithm and SGBM improved al-gorithm.Simple background simulation results show that the measurement error of the untrimmed BM optimization al-gorithm is less than 0.9%,and the simulation time is 2s.Compared with other algorithms,the simulation time is the shortest and the real-time performance is the highest.Complex background simulation results show that the measure-ment error of the trimmed SGBM improved algorithm is 0.4%,and the simulation times is 2.5 s.On the basis of im-age prior information,this paper improves the real-time performance of stereo matching by optimizing the BM algo-rithm,and improves the accuracy of stereo matching based on the improved SGBM algorithm,which provides a theo-retical basis for the practical application of binocular vision system.
维普
万方数据