双目视觉图像边缘区域处理的精确立体匹配  

Accurate Stereo Matching for Processing Binocular Vision Image Edge Region

作　　者：符强[1,2,3] 孔健明 纪元法 任风华[1,2,3] FU Qiang;KONG Jian-ming;JI Yuan-fa;REN Feng-hua(Guangxi Key Laboratory of Precision Navigation Technology and Application,Guilin University of Electronic Technology,Guilin Guangxi 541004,China;Information and Communication School,Guilin University of Electronic Technology,Guangxi Guilin 541004,China;National&LocalJoint Engineering Research Center of Satellite Navigation Positioning and Location Service,Guilin Guangxi 541004,China)

机构地区：[1]桂林电子科技大学广西精密导航技术与应用重点实验室,广西桂林541004 [2]桂林电子科技大学信息与通信学院,广西桂林541004 [3]卫星导航定位与位置服务国家地方联合工程研究中心,广西桂林541004

出　　处：《计算机仿真》2023年第12期226-231,共6页Computer Simulation

基　　金：国家自然科学基金(61561016,61861008);广西科技厅项目(桂科AA1918200);“认知无线电与信息处理”教育部重点实验室(CRKL200108);广西精密导航技术与应用重点实验室(DH201901);桂林电子科技大学研究生教育创新计划项目(2022YCXS050)。

摘　　要：立体匹配是双目视觉系统最关键一步,其处理图像的速度和精度直接影响系统结果。针对传统SAD(Sum of absolutedifferences)算法存在较多冗余计算问题,引入线性插值法降低算法复杂度;针对光照对算法干扰,提出一种融合算法策略避免光照影响;为了提高融合算法处理图像精度,引入Sobel边缘检测算子及自适应惩罚系数约束和保护图像边缘区域;为了获取边缘信息且抵御光照变化影响,采用基于结构森林快速边缘检测及大律法(OTSU)进行图像边缘和非边缘区域的分割;针对图像边缘出现的遮挡,采用左右一致性检测(Left-rightchecking)算法,并采用引导滤波器实现边缘优化。将该算法与传统SGM(Semi-global matching)算法在Middlebury测试集进行对比实验,平均误匹配率降到7.55%以下,平均运算时间较SGM算法缩短3.5s。Stereo matching is the most important step in binocular vision systems,and the speed and precision of image processing directly affect the results of the system.To solve the problem of redundancy in the traditional SAD algorithm,linear interpolation is introduced to reduce the algorithm complexity.Aiming at the interference of illumi⁃nation,a fusion algorithm strategy is proposed to avoid the influence of illumination.In order to improve the image processing accuracy of the fusion algorithm,the Sobel edge detection operator and adaptive penalty coefficient con⁃straint are introduced to protect the image edge region.In order to obtain the edge information and resist the influence of light changes,a structured forest based fast edge detection and large law method is used to segment the edge and non-edge regions of the image.The left-right checking algorithm is used for the occlusion of the image edge,and the bootstrap filter is used to optimize the edge.Compared with the traditional semi-global matching algorithm in the Middlebury test set, the average mismatching rate is less than 7. 55%, and the average computing time is shortenedby 3. 5s compared with the SGM algorithm.

关 键 词：双目视觉 立体匹配 线性插值 自适应惩罚系数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]