基于截断AD-Census函数的立体匹配算法  

Stereo Matching Algorithm Based on Truncated AD Census Function

作　　者：张健[1] 覃翠[1] ZHANG Jian;QIN Cui(School of Information and Communication Engineering,Nanjing Institute of Technology,Nanjing Jiangsu 211167,China)

机构地区：[1]南京工程学院信息与通信工程学院,江苏南京211167

出　　处：《电子器件》2023年第1期109-114,共6页Chinese Journal of Electron Devices

基　　金：江苏省自然科学青年基金项目(BK20191012)。

摘　　要：双目视觉技术是重要的立体成像技术之一,该技术的核心是计算匹配像素点之间的视差值,而这种计算依赖于代价函数的设计。其中Absolute Difference Census(AD-Census)函数作为代价函数是常用的方法,但该方法受权重系数的影响较大。为了有效提高AD-Census函数的匹配精度,提出向AD-Census函数中添加截断参数以改进其匹配效果,保证了代价聚合时像素的绝对值和梯度不会出现因其中一个值较大,而导致另一个值的影响被忽略的情况出现,从而提高匹配精度,与传统的AD-Census函数相比失配率可以降低6%以上。Binocular vision technology is one of the important stereo imaging technologies.The core of this technology is to calculate the disparity between matching pixels,which relies on the design of cost function.The absolute difference census(AD-census)function is commonly used as the cost function.However,this function is greatly affected by the weight coefficients.In order to improve the matc-hing accuracy of AD-census function,truncation parameters are proposed to be added to AD-census function to improve the matching precision,so as to ensure that the absolute value and gradient of pixels will not be ignored due to the large value of one of them.Com-pared with the traditional AD-census function,the mismatch rate can be reduced by more than 6%.

关 键 词：截断AD-Census函数 立体匹配 代价函数 匹配精度 误匹配率 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]