Analysis and performances of an object localization and a dimensional measurement method applied to the calibration of cameras

The calibration of a camera is an indispensable step when high precision vision is to be used as a measuring instrument. It consists in the determination of the intrinsic and extrinsic parameters of the vision system using a reference test pattern. To do this, one needs to localize and position the elements of the test pattern in the digital image with a maximum of accuracy. The objective is to obtain a resolution smaller than the dimensions of a pixel. In this context, it is necessary to implement "subpixel" precision algorithms for the localization of objects obtained by an edge detection approach. In this work, the authors have noted that the edge points are not detected in all the directions when the image is interpolated using orthogonal polynomials according to Haralick's method. Other methods have also been studied such as Canny Deriche or the spline approach. Finally, the authors have developed a method using a regression polynomial of third degree. They define, as the edge, the locus of the points for which the directional derivative is maximal. This locus is also the locus for which the second directional derivative is equal to zero. They have studied the localization and dimensional errors obtained using a test pattern consisting of circles as well as the influence of the dimension and the position of the test pattern, and of the noise and blur effects on the error. The results are analyzed and discussed.