Vision measurements are becoming a powerful technique for operational modal analysis. Nevertheless, the performance of the existing motion extraction algorithms is subject to perturbations due to environmental interference in practical applications, such as illumination variations and objects in the background of the investigated structure. This paper applies a novel image-based motion estimation method that explicitly considers such perturbations, and compares the performance of different motion extraction methods with regards to the modal parameter estimates. These methods are tested on video data recorded from a cantilever beam in the laboratory under random base excitation, where the robustness against ambient light changes and disturbance by background features is investigated. The goal of this work is to move towards real-world applications of vision-based operational modal analysis by improving robustness.