Estimation of extrinsic parameters for dynamic binocular stereo vision using unknown-sized rectangle images.

An online calibration method of extrinsic parameters for dynamic binocular stereo vision using unknown-sized rectangle images is presented herein. The two cameras can rotate freely in two-axis direction without zooming, and the cameras' intrinsic parameters and initial relative position are known in advance. Hence, only the rotation angles of the cameras, namely, the yaw angle in the horizontal direction and the pitch angle in the vertical direction, must be solved. The approximate values of the rotation angles are obtained by the Ellipse-A attitude and heading reference system in real time, and the rotation angles of each camera are solved iteratively using only a single rectangular centroid according to interimage homography. The pitch and yaw angles are eventually modified based on the property of fundamental matrix by the corresponding rectangular centroids, obtained by the left and right cameras after rotation, respectively. Computer simulations and real experiments are performed to validate the efficiency of the proposed method, and we compare calibration results with that of the scale-invariant feature transform method and Zhang's method. The quantitative results demonstrate that the root mean square (rms) of the absolute error of the attitude angles and the translation vector does not exceed 0.12° and 7.2 mm, respectively. Additionally, the rms of the relative error of the three-dimensional coordinates does not exceed 1.6%.