基于改进局部一致性约束的立体匹配算法
Stereo matching algorithm based on improved local consistency constraint
辽宁工程技术大学 电子与信息工程学院,辽宁 葫芦岛 125105
图 & 表
图 1. 本文算法流程图
Fig. 1. Algorithm flow chart of this paper
下载图片 查看原文
图 2. 三角剖分步骤
Fig. 2. Triangulation step
下载图片 查看原文
图 3. 局部一致性与随机倾斜窗口对比
Fig. 3. Comparison of local consistency and random window
下载图片 查看原文
图 4. 不同算法在Middlebury 3.0图像对上的实验结果。(a)参考图像;(b)真实视差图;(c)PMS算法;(d)本文算法。
Fig. 4. Lifting effect comparison charts of different algorthms on Middlebry 3.0 image pairs.(a)Reference image;(b)Truth graph;(c)PMS algorithm;(d)Proposed algorithm.
下载图片 查看原文
图 5. 本文提出的两种模型对比
Fig. 5. Comparison of the two models proposed in this paper
下载图片 查看原文
图 6. 不同算法在Middlebury3.0图像对上的实验结果。(a)参考图像;(b)真实视差图;(c)AD-Census算法;(d)ELASF算法;(e)GA-Net算法;(f)本文算法。
Fig. 6. Experimental results of different algorithms on Middlebury 3.0 image pairs.(a)Reference image;(b)Real disparity map;(c)AD-Census algorithm;(d)ELASF algorithm;(e)GA-Net algorithm;(f)Algorithm in this paper.
下载图片 查看原文
图 7. 本文方法与对比算法在非遮挡区域的对比
Fig. 7. Comparison of the method in this paper and the comparison algorithm in the non-occluded area
下载图片 查看原文
图 8. 本文方法与对比算法在全部区域的对比
Fig. 8. Comparison of the method in this paper and the comparison algorithm in all areas
下载图片 查看原文
图 9. 本文算法与PMS算法的11次迭代结果
Fig. 9. Results of 11 iterations of the algorithm in this paper and the PMS algorithm
下载图片 查看原文
表 1本文算法与PMS算法结果对比
Table1. Comparison of the results of the algorithm in this paper and the PMS algorithm
Data name | non/% | all/% | Eavg/% |
---|
PMS | Prop-sla | Prop-Par | PMS | Prop-sla | Prop-Par | PMS | Prop-sla | Prop-Par |
---|
Adirondack | 22.09 | 21.27 | 6.669 | 24.68 | 23.61 | 10.41 | 1.93 | 1.9 | 0.63 | ArtL | 15.48 | 14.14 | 13.25 | 25.42 | 24.82 | 27.23 | 1.36 | 1.23 | 1.19 | Jadeplant | 24.95 | 24.93 | 45.66 | 39.85 | 39.32 | 55.79 | 4.08 | 4.34 | 15.36 | Motorcycle | 14.49 | 13.73 | 11.67 | 21.68 | 21.09 | 18.6 | 1.12 | 1.09 | 1.01 | MotorcycleE | 19.93 | 17.51 | 13.92 | 27.22 | 24.64 | 20.84 | 1.4 | 1.29 | 1.18 | Piano | 27.03 | 25.95 | 25.67 | 34.95 | 33.52 | 33.16 | 1.59 | 1.46 | 1.45 | PianoL | 43.56 | 43.82 | 34.13 | 50 | 48.8 | 41.95 | 3.18 | 3.11 | 2.33 | Pipes | 17.76 | 16.94 | 15.39 | 30.97 | 30.1 | 28.55 | 2.03 | 1.88 | 2.01 | Playroom | 31.99 | 30.75 | 27.15 | 42.77 | 41.46 | 39.04 | 2.91 | 2.82 | 2 | Playtable | 19.53 | 19.8 | 17.54 | 25.94 | 26.46 | 24.89 | 1.19 | 1.25 | 1.13 | PlaytableP | 12.48 | 11.9 | 10.7 | 18.7 | 18.25 | 17.05 | 0.65 | 0.64 | 0.54 | Recycle | 21.42 | 21.3 | 16.65 | 25.11 | 25.4 | 19.72 | 1.57 | 1.59 | 1.01 | Shelves | 49.86 | 49.76 | 46.63 | 56.34 | 56.11 | 52.82 | 4.59 | 4.55 | 3.23 | Teddy | 6.809 | 7.609 | 6.037 | 14.81 | 14.46 | 13.91 | 0.64 | 0.65 | 0.48 | Vintage | 29.7 | 29.51 | 40.26 | 34.67 | 34.05 | 44.05 | 2.15 | 2.11 | 4.67 |
|
查看原文
表 2对比算法在不同区域的平均误匹配率
Table2. Average mismatch rates of algorithms in different regions
Ebad/% | AD-census | ELASF | GA-net | Proposed |
---|
non | 26.19 | 22.23 | 26.91 | 21.89 | all | 34.59 | 30.53 | 30.17 | 29.71 |
|
查看原文
表 3不同迭代次数的误匹配率
Table3. Mismatch rate for different iterations
Data name | 第一次迭代 | 第2次迭代 | 第3次迭代 |
---|
PMS | Prop-sla | Prop-par | PMS | Prop-sla | Prop-par | PMS | Prop-sla | Prop-par |
---|
Adirondack | 20.08 | 16.43 | 9.820 | 21.35 | 20.37 | 6.574 | 22.09 | 21.27 | 6.669 | ArtL | 18.35 | 15.48 | 22.03 | 15.42 | 14.43 | 13.54 | 15.48 | 14.14 | 13.25 | Jadeplant | 28.23 | 27.58 | 49.50 | 25.25 | 24.84 | 46.54 | 24.95 | 24.93 | 45.66 | Motorcycle | 15.82 | 14.19 | 13.72 | 14.68 | 13.78 | 11.44 | 14.49 | 13.73 | 11.67 | MotorcycleE | 21.26 | 17.59 | 16.08 | 20.30 | 17.41 | 13.57 | 19.93 | 17.51 | 13.92 | Piano | 28.43 | 26.92 | 28.41 | 27.19 | 25.95 | 25.88 | 27.03 | 25.95 | 25.67 | PianoL | 43.32 | 40.45 | 37.01 | 43.79 | 42.37 | 34.41 | 43.56 | 43.82 | 34.13 | Pipes | 18.51 | 16.82 | 19.55 | 17.79 | 16.84 | 15.58 | 17.76 | 16.94 | 15.39 | Playroom | 33.01 | 30.22 | 31.27 | 32.12 | 30.48 | 27.07 | 31.99 | 30.75 | 27.15 | Playtable | 17.99 | 18.45 | 22.34 | 19.03 | 19.52 | 17.59 | 19.53 | 19.80 | 17.54 | PlaytableP | 12.45 | 11.75 | 14.65 | 12.43 | 12.07 | 10.90 | 12.48 | 11.90 | 10.70 | Recycle | 21.66 | 20.45 | 18.85 | 20.76 | 20.87 | 16.54 | 21.42 | 21.30 | 16.65 | Shelves | 50.53 | 50.10 | 46.20 | 50.09 | 49.76 | 46.29 | 49.86 | 49.76 | 46.63 | Teddy | 7.471 | 7.015 | 9.251 | 6.83 | 7.632 | 6.013 | 6.809 | 7.609 | 6.097 | Vintage | 31.64 | 30.20 | 45.19 | 29.78 | 29.12 | 41.14 | 29.70 | 29.51 | 40.26 |
|
查看原文
任晓奎, 关钧渤, 殷新勇, 陶志勇. 基于改进局部一致性约束的立体匹配算法[J]. 液晶与显示, 2023, 38(4): 543. Xiao-kui REN, Jun-bo GUAN, Xin-yong YIN, Zhi-yong TAO. Stereo matching algorithm based on improved local consistency constraint[J]. Chinese Journal of Liquid Crystals and Displays, 2023, 38(4): 543.