2014 , Volume 19, № 4, p.61-68

Most of algorithms in 3-D Computer Vision employ chessboard pattern recognition for camera calibration because of its . However, such algorithms often use low effective chessboard feature detection based on selection of threshold value or its range. In doing so, the robustness of grid points recognition and performance are decreased. To increase the robustness, we propose to use statistical descriptors like mean, variance and skewness of the shape of the response function of local regions at the image, which contains feature points. Consequently, one can decrease the number of false "grid points" that should be analyzed at the chessboard's image. Full algorithm consists of the traditional Harris detection (detector) followed by additional filtration, which uses above mentioned statistical descriptors in small regions centered at local maximums. Then all recognized grid points are to be analyzed with a higher rate to find out ones that really belong to the chessboard (because that approach has less error value). The obtained experimental results show that the offered algorithm with the introduced additional filtration procedure leads to better performance compared with the recently available algorithms. The robustness was increased up to 85 % at the expense of a minor 15 % increase of the computational time.

Aknowlegements: This research is supported by the SB RAS project III.23.1.3, RFBR grants 13-07-00009 and the Grant of President of Russian Federation SP-816.2012.5.

Received 4 February 2014.