基于图像分割的端到端双目立体匹配算法研究  

Research on End-to-end Binocular Stereo Matching Algorithm Based on Image Segmentation

作　　者：唐明[1] 殷磊 白俊卿 TANG Ming;YIN Lei;BAI Junqing(School of Intelligent Science and Engineering,Xi'an Peihua University,Xi'an 710199;School of Computer Science,Xi'an Shiyou University,Xi'an 710065)

机构地区：[1]西安培华学院智能科学与工程学院,西安710199 [2]西安石油大学计算机学院,西安710065

出　　处：《计算机与数字工程》2024年第12期3756-3760,共5页Computer & Digital Engineering

摘　　要：针对立体匹配算法因为光照不均匀、遮挡等因素出现的低纹理、重复纹理的病态区域从而导致参数量大、精度低等问题,提出一种基于图像分割辅助引导的端到端双目立体匹配优化算法。该算法将利用高斯滤波预处理和超像素分割算法得到的图像区域作为StereoNet网络的输入,经过代价计算、代价聚合、视差计算以及视差优化等过程,有效地适配于双目立体匹配算法中,同时对视差估计过程进行优化,实现了目标图像在低分辨率特征下高效率的立体视差估计。并利用Middlebury测试平台标准数据进行验证,验证结果表明论文算法在病态区域中能够减少内存消耗,提高运行速度,提升视差图边缘区域的辨识度。该立体匹配算法具备一定的工程指导意义。Aiming at the ill conditioned areas of low texture and repeated texture in stereo matching algorithm due to uneven il-lumination and occlusion,resulting in large amount of parameters and low accuracy,an end-to-end binocular stereo matching opti-mization algorithm based on image segmentation auxiliary guidance is proposed.The algorithm takes the image area obtained by Gaussian filter preprocessing and super-pixel segmentation algorithm as the input of stereonet network.After the process of cost cal-culation,cost aggregation,parallax calculation and parallax optimization,it is effectively adapted to the binocular stereo matching algorithm.At the same time,the parallax estimation process is optimized to realize the efficient stereo parallax estimation of the tar-get image under the characteristics of low resolution.The standard data of Middlebury test plafform are used to verify.The verifica-tion results show that the algorithm in this paper can reduce memory consumption,improve running speed and improve the identifi-cation of parallax edge region in ill conditioned region.The stereo matching algorithm has certain engineering guiding significance.

关 键 词：双目视觉 图像分割 高斯滤波 StereoNet 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]