摘要
针对局部立体匹配算法在弱纹理区域匹配精度低的问题,提出一种基于扫描线优化的多特征融合立体匹配算法。首先,在代价计算阶段采用自适应权重SAD代价、梯度代价和Census代价融合进行代价计算,得到更加准确的多特征融合代价;然后,采用自适应窗口代价聚合的方法对局部窗口内的初始匹配代价进行聚集;再采用扫描线优化的方法对聚合代价进行优化;最后,采用胜者为王的策略进行视差计算,并采用一致性检查、迭代局部投票、视差填充、视差非连续区域调整、子像素优化的方法进行视差优化得到最终的视差图。实验结果表明,在cone和teddy两组图像的测试中加入扫描线优化方法后,弱纹理区域的匹配效果得到明显改善,匹配精度分别提升3.19%和1.21%;为了验证本文算法的综合性能,在cone、teddy、tusakub、venus四组图像中进行测试并与其它算法进行对比,在算法对比实验中,所提算法获取的视差图误匹配点最少,平均误匹配率为5.64%,平均运行时间为3.12s,能够在保证较高匹配精度的同时拥有较快的运算速度,具有较好的综合性能。
Aiming at the problem that local stereo matching algorithms have low matching accuracy in weak texture regions,a multi-feature fusion stereo matching algorithm basedon scanline optimization is proposed.First,in the cost calculation stage,adaptive weight SAD cost,gradient cost,and Census cost fusion are used for cost calculation to obtain a more accurate multi feature fusion cost;Then,the adaptive window cost aggregation method is used to aggregate the initial matching costs within the local window;Further use the scanning line optimization method to optimize the aggregation cost;Finally,the winner is the king strategy is adopted for disparity calculation,and methods such as consistency check,iterative local voting,disparity filling,adjustment of non continuous disparity regions,and sub-pixel optimization are used for disparity optimization to obtain the final disparity map.The experimental results show that the addition of scan line optimization method in the testing ofcone and teddy images significantly improves the matching effect of weak texture areas,with matching accuracy increased by 3.19%and 1.21%respectively;In order to verify the comprehensive performance of the algorithm proposed in this paper,tests were conducted on four sets of images:cone,teddy,tusakub,and venus,and compared with other algorithms.In the algorithm comparison experiment,the algorithm obtained the least disparity map mismatch points,with an average mismatch rate of 5.64%and an average running time of 3.12 seconds.It can ensure high matching accuracy while having a fast computational speed and good comprehensive performance.
作者
赵增旭
刘向阳
任彬
胡连庆
ZHAO Zeng-xu;LIU Xiang-yang;REN Bin;HU Lian-qing(College of Mechanical Engineering,Shijiazhuang Tiedao University,Shijiazhuang Hebei O50043,China)
出处
《计算机仿真》
2025年第11期336-341,354,共7页
Computer Simulation
关键词
立体匹配
多特征融合
自适应窗口代价聚合
扫描线优化
Stereo Matching
Multi-feature fusion
Adaptive window cost aggregation
Scan line optimization
