Combined evolution strategies for dynamic calibration of video-based measurement systems

Calibration is a crucial step to obtaining three-dimensional (3-D) measurement using video camera-based stereo systems. Approaches based on epipolar geometry are particularly appealing as there is no need to know the 3-D position of the control points a priori and because the solution is found by solving a set of linear equations through matrix manipulation. Indeed, all the parameters can be determined except for the pair of principal points, which poses a considerable drawback. Whereas in low-accuracy systems (two-dimensional measurement error < 0.2 pixels) such points can be assumed to lie at the image center without degrading the overall 3-D accuracy, in high-accuracy systems their true position must be computed accurately. In this case, all the calibration parameters (including the principal points) can still be estimated through epipolar geometry, but it is necessary to minimize a highly nonlinear cost function. It is shown here that by combining two evolutionary optimization strategies this minimization can be carried out, both efficiently (in quasi-real time) and reliably (avoiding local minima). The resulting strategy, which we call enhanced evolutionary search (EES), allows the full calibration of a stereo system using only a rigid bar; this simplicity is a definite step forward in stereo-camera calibration. Moreover, EES can be applied to a wide range of applications where the cost function contains complex nonlinear relationships among the optimization variables.