In modern autonomous systems, measurement repeatability and precision are crucial for robust decision-making algorithms. Stereovision, which is widely used in safety applications, provides information about the object's shape, orientation, and localisation in a 3D space. The camera's lens distortion is a common factor that introduces measurement errors.
Traditional calibration methods require a test for each camera to determine the possible correction parameters. However, this method is unfeasible for large-scale production due to the time consumption and complexity. In this paper, a general correction model is developed using a statistical approach to minimise the effect of lens distortion across different cameras of the same type: Basler Lenses (C125-0618-5M F1.8 f6mm) assembled with Sony IMX477R matrices.
This is done with the aid of a novel method for lens distortion measurement based on linear regression with photographed vertical and horizontal lines. The measure for the general correction model is 3.0 and 4.5, for horizontal and vertical lines respectively, while individual scores of cameras range from 1.5 to 7.8 for horizontal, and 1.2 to 23.4 for vertical lines. Furthermore, the lens distortion correction model is validated in stereovision applied to bird tracking around wind farms. The correction of synthetically generated bird's flight trajectories can reduce the error of around 15-20\% in disparity and depth estimation in certain regions of the image, e.g. a bird at a distance of 610 meters (5 pixel disparity) is seen by the distorted lens at 520 meters (5.9 pixel disparity). The results ensure that the presented general correction model meets the accuracy requirements of multi-camera applications.