Robust and Accurate Calibration Point Extraction with Multi-scale Chess-Board Feature Detector - Advances in Image and Graphics Technologies - page 159

Fig. 10. Bidirectional growth algorithm

The lines successfully fit are illustrated in Fig. 11 .

Fig. 11. Results of line fitting

The four ID codes in the corners of the grid are different from each other and have
quite stable and distinctive appearances that are easy to be recognized. The 3D coor-
dinates of the vertices detected are then obtained by using the method in [15].

4 Experimental Results

To show the contributions of our approach, we have performed the following two
experiments: First, to evaluate the precision we compare the average reprojection error
of the chess-board vertices detected by our detector with that by Harris and ChEss;
Second, to evaluate the robustness we compare the chess-board recognition success
rate of our method with that of J. Sun [16] and de la Escalera. We conduct an additional
experiment over the images with different illumination, image sizes and camera poses
to prove the validity of our method.

4.1 Precision

The dataset we use for this experiment consists of 20 groups (each group has about 20
images captured around the object on the chess-board) of images with good illumination.