Robust and Accurate Calibration Point Extraction with Multi-scale Chess-Board Feature Detector - Advances in Image and Graphics Technologies

Graphics Reference

In-Depth Information

These images can also be divided into 3 categories according to the angle between the

image plane and the chess-board plane: high position (the angle is less than 30 degree),

middle position (the angle is less than 60 degree and more than 30 degree), low position

(the angle is more than 60 degree). As stated before, the camera focuses mainly on the

objects to be reconstructed, so the calibration pattern is often captured with imperfect

focus, which often cause blurriness in the condition of low position. We separately

measure the average distance between the projected points of the 3D vertices and the

vertices detected with Harris, ChEss and our detector. As shown in Fig. 13 , our detector

performs as well as ChEss, while much better than Harris, the average reprojection error is

less than 1 pixel in all three conditions. We can also see that imperfect focus affects the

accuracy, but our detector performs much more stable than Harris. ChEss is rather accu-

rate but tends to produces much more false features and the sing-scale property also ag-

gravates its limitation.

(a) (b) (c)

Fig. 12. Three categories divided by camera positions (a) high position (b) middle position

(c) low position

/ 通用格式

low position

middle position

high position

/ 通用格式

our detector

ChEss

Harris

Fig. 13. Reprojection error of ChEss, Harris and our detector

Successful chess-board detection is to extract enough vertices, while discarding the

false ones, and map them to 3D locations in the board to complete the camera calibration.

There are generally three kinds of approaches in this field: J. Sun's method using LSC

and line fitting scheme, Escalera method combining Harris with Hough transformation.

Search WWH ::

Custom Search

Home