基于区域先验信息的去遮挡立体匹配算法 下载: 1270次
De-Occlusion Stereo Matching Algorithm Based on Regional Prior Information
陈先锋 1,2,3郭正华 1,2,3伍俊龙 1,2,3马帅 1,2,3杨平 1,2许冰 1,2,*
1 中国科学院光电技术研究所自适应光学重点实验室, 四川 成都 610209
2 中国科学院光电技术研究所, 四川 成都 610209
3 中国科学院大学, 北京 100049
图 & 表
图 1. 代价计算
Fig. 1. Cost calculation
下载图片 查看原文
图 2. 视差精化示意图 (a)左右一致性检测;(b)区域一致性检测
Fig. 2. Disparity refinement diagrams. (a) Left-right consistency check; (b) regional consistency check
下载图片 查看原文
图 3. 遮挡距离解释图。(a)参考图; (b)目标图
Fig. 3. Occlusion distance interpretation diagrams. (a) Reference image; (b) target image
下载图片 查看原文
图 4. 仿真展示图。(a)左图;(b)真值图;(c) NCC残差图;(d)本文算法残差图
Fig. 4. Simulation results. (a) Left images; (b) ground truths; (c) residual figures of NCC; (d) residual figures of proposed algorithm
下载图片 查看原文
表 1左右图存在大量遮挡的误匹配率
Table1. Mismatching rates of left and right figures with large occlusion%
Item | NCC | Proposed method |
---|
Wood1 | 7.73 | 4.59 | Baby1 | 6.08 | 4.07 | Baby2 | 9.32 | 7.98 | Lampshade1 | 20.00 | 16.23 | Lampshade2 | 20.99 | 17.39 | Monopoly | 30.11 | 28.21 |
|
查看原文
表 2左右图存在较少遮挡的误匹配率
Table2. Mismatching rates of left and right figures with less occlusion%
Item | NCC | Proposed method |
---|
Aloe | 12.14 | 11.12 | Bowling1 | 16.72 | 15.39 | Cloth1 | 3.17 | 2.83 | Cloth2 | 9.62 | 8.66 | Rocks1 | 10.48 | 10.26 | Wood2 | 9.33 | 9.12 |
|
查看原文
表 3原始算法与引入本文算法后的误匹配率
Table3. Mismatching rates of original algorithm and proposed algorithm%
Item | NCC | NCC combined withproposed method | SAD | SAD combined withproposed method | SSD | SSD combined withproposed method |
---|
Wood1 | 7.73 | 4.59 | 17.80 | 13.78 | 15.84 | 12.28 | Baby1 | 6.08 | 4.07 | 18.33 | 15.18 | 13.64 | 11.33 | Baby2 | 9.32 | 7.98 | 23.96 | 21.08 | 21.92 | 19.70 | Lampshade1 | 20.00 | 16.23 | 43.23 | 39.45 | 39.71 | 35.60 | Lampshade2 | 20.99 | 17.39 | 58.58 | 57.85 | 53.80 | 52.76 | Monopoly | 30.11 | 28.21 | 53.18 | 50.12 | 50.37 | 47.38 |
|
查看原文
陈先锋, 郭正华, 伍俊龙, 马帅, 杨平, 许冰. 基于区域先验信息的去遮挡立体匹配算法[J]. 激光与光电子学进展, 2019, 56(19): 191001. Xianfeng Chen, Zhenghua Guo, Junlong Wu, Shuai Ma, Ping Yang, Bing Xu. De-Occlusion Stereo Matching Algorithm Based on Regional Prior Information[J]. Laser & Optoelectronics Progress, 2019, 56(19): 191001.