摘要
针对传统立体匹配算法匹配窗口采用固定阈值以及代价聚合采用固定权重,使得算法在边界区域和弱纹理区域匹配精度低的问题,提出一种基于自适应窗口与权重的立体匹配算法。该算法首先根据图像像素梯度信息自动生成颜色阈值,进而创建十字支撑窗口,并对窗口像素点进行相似度计算,按照3-σ原则剔除干扰像素,然后用当前窗口最小臂长和臂长阈值自动计算灰度差绝对值之和和Census的权重,利用指数归一化函数聚合SAD和Census代价,通过win-take-all算法策略进行初始视差计算,最后通过左右视差一致性原则筛选和滤波得到精密视差图。使用改进算法在Middlebury测速平台对标准图像进行实验测试,平均错误率在4.5%以内,较原有算法错误率降低了11.5%左右。
Aiming at the problem that the traditional stereo matching algorithm uses fixed threshold for the matching window and fixed weight for the cost aggregation,which makes the algorithm have low matching accuracy in the boundary region and weak texture region,this paper presents a stereo matching algorithm based on adaptive window and weight.The algorithm firstly based on image pixel gradient information automatically generated color threshold,and then creates windows cross bracing,and carries out similarity calculation on the window pixels,eliminates interference in accordance with the principle of 3-σpixels,and then uses the forearm length and arm length threshold in the current window to automatically calculate weight of the sum of absolute differences(SAD)and Census,using aggregate index normalized function SAD and Census costs,through the windows-take-all strategy calculates initial parallax,finally by principle of horizontal parallax the consistency sifts and filters precision parallax figure.The improved algorithm is used to test the standard image on the middlebury speed measurement platform,the average error rate is within 4.5%,which is about 17.4%lower than the original algorithm.
作者
路乾坤
李彦
Lu Qiankun;Li Yan(School of Electronics Information,Jiangsu University of Science and Technology,Zhenjiang 212003,China)
出处
《电子测量技术》
2019年第14期117-122,共6页
Electronic Measurement Technology
关键词
立体匹配
自适应权重
自适应窗口
颜色阈值
干扰值剔除
stereo matching
adaptive weight
adaptive window
color threshold
interference value rejection
