基于单张mark点图片的畸变校正算法

100视觉币有点太高了，没有这么多视觉币怎么办？

18352405919 发表于 2024-5-20 20:11
参考网上有篇帖子 吊炸天的自标定,就是用一张图做矫正

他怎么区分镜头原因还是角度原因？

分割出四个角点做投影应该就差不多了吧，尝试一下:

1. read_image (Image, 'D:/标定板图像畸变校正.png')
   
2. rgb1_to_gray (Image, GrayImage)
   
3. 
   
4. binary_threshold (GrayImage, Region, 'max_separability', 'dark', UsedThreshold)
   
5. connection (Region, ConnectedRegions)
   
6. 
   
7. select_shape (ConnectedRegions, SelectedRegions, 'circularity', 'and', 0.65, 1)
   
8. 
   
9. select_shape_std (SelectedRegions, MaxRegions, 'max_area', 70)
   
10. region_features (MaxRegions, 'max_diameter', MaxDiameter)
    
11. 
    
12. region_features (SelectedRegions, 'row', RowX)
    
13. region_features (SelectedRegions, 'column', ColX)
    
14. MinRow := min(RowX)
    
15. MinCol := min(ColX)
    
16. MaxRow := max(RowX)
    
17. MaxCol := max(ColX)
    
18. 
    
19. select_shape (SelectedRegions, SelectedRegionsTL, ['row', 'column'], 'and', [0,0], [MinRow+1.5*MaxDiameter, MinCol+1.5*MaxDiameter])
    
20. select_shape (SelectedRegions, SelectedRegionsTR, ['row', 'column'], 'and', [0,MaxCol-1.5*MaxDiameter], [MinRow+1.5*MaxDiameter,99999])
    
21. select_shape (SelectedRegions, SelectedRegionsBL, ['row', 'column'], 'and', [MaxRow-1.5*MaxDiameter,0], [99999,MinCol+1.5*MaxDiameter])
    
22. select_shape (SelectedRegions, SelectedRegionsBR, ['row', 'column'], 'and', [MaxRow-1.5*MaxDiameter,MaxCol-1.5*MaxDiameter], [99999,99999])
    
23. 
    
24. area_center (SelectedRegionsTL, Area, RowTL, ColTL)
    
25. area_center (SelectedRegionsTR, Area, RowTR, ColTR)
    
26. area_center (SelectedRegionsBL, Area, RowBL, ColBL)
    
27. area_center (SelectedRegionsBR, Area, RowBR, ColBR)
    
28. Px := [RowTL, RowTR, RowBR, RowBL]
    
29. Py := [ColTL, ColTR, ColBR, ColBL]
    
30. 
    
31. distance_pp (RowTL, ColTL, RowTR, ColTR, Distance)
    
32. Qx := [RowTL, RowTL, RowTL+Distance, RowTL+Distance]
    
33. Qy := [ColTL, ColTL+Distance, ColTL+Distance, ColTL]
    
34. 
    
35. hom_vector_to_proj_hom_mat2d (Px, Py, [1,1,1,1], Qx, Qy, [1,1,1,1], 'normalized_dlt', HomMat2D)
    
36. 
    
37. projective_trans_image (GrayImage, TransImage, HomMat2D, 'bilinear', 'true', 'true')

复制代码

我这个支持任意畸变校正

freekun139com 发表于 2024-5-31 14:38
分割出四个角点做投影应该就差不多了吧，尝试一下:
read_image (Image, 'D:/标定板图像畸变校正.png')
rgb1 ...

肯定不行啊

halcon2016 发表于 2024-5-31 14:49
肯定不行啊

确实不行，还是有畸变的

很厉害

好強阿 來學習一下了讚

freekun139com 发表于 2024-5-31 14:38
分割出四个角点做投影应该就差不多了吧，尝试一下:

这个和Opencv的那个原理很像

这张mark图哪来的？