Medical robotics have revolutionized medical practice, providing precision and stability in surgical procedures and reducing the risk of human error. Future technological innovations of this branch require ever-greater strides in the direction of accuracy and reliability of every part of the procedure in order to minimize irregularities and provide the best possible service to patients. As part of this work, an algorithm for robust localization of retro-reflective markers using a stereo-vision system used during the preparation for robotic neuronavigation was thoroughly tested and optimized. Markers are used for preoperative registration of the patient in physical space. In order to cover all required scenarios, a laboratory setup was set up in the Medical Robotics Laboratory of the Regional Center of Excellence for Robotic Technologies, where the camera is separated from the robot and placed in front of a plastic phantom that serves as a replica of a human head, with an anthropomorphic appearance and similar light reflection to human skin. The application underwent systematic testing, examining its functionality across all possible positions in relation to the spheres, consistent with previous usage. By calibrating the used camera, the correctness of the displayed image was verified, based on which the more advanced algorithm of adaptive segmentation of the required structures in the image proved to be an obvious step towards more efficient localization of the positions of required markers. The application was developed in the C++ programming language with the help of the OpenCV computer vision library.