Robust and Accurate Calibration Point Extraction with Multi-scale Chess-Board Feature Detector - Advances in Image and Graphics Technologies - page 161

Fig. 14. An example where Harris corner deviants from true vertices

Our method differs with these two approaches. The comparison will be conducted in this
experiment.
The dataset we use here consists of 40 groups of images which can also be divided
into three categories like 4.1. We measure the chess-board recognition success rate and
show some examples where these three methods would fail.
Fig. 15 shows that our method succeeds to calibrate the most images, which means
it's the most robust among the four methods. The other three perform rather badly in the
condition of low camera position. The line fitting scheme of de la Escalera is at a
disadvantage when distortion or blurring exists. J. Sun's method has difficulty with
discarding false vertices.

/ 通用格式

/ 通用格式

/ 通用格式

low position
middle position
high position

/ 通用格式

/ 通用格式

/ 通用格式

Our
method

J.Sun Escalera

Fig. 15. Chess-board recognition success rates

4.2 Other Factors

We conduct this experiment on 200 images taken by us with different illumination,
image sizes and camera poses and another 200 images taken by volunteers with little
knowledge about our algorithm except the suggestion of capturing the object and the
board at the same time. As shown in Fig. 16 , we successfully complete calibration of