Accurate and stable feature tracking is an important prerequisite for visual localization. In particular, methods of constructing cooperative targets based on optical beacons with prior feature called optical cooperative localization are drawing increasing attention and application. In this work, we propose a robust beacon tracking method based on improved kernel correlation filters for optical cooperative localization. Comprehensive feature verification in terms of area, contour, geometry and matching is used to effectively improve the robustness of beacon tracking in practical application scenarios while ensuring real-time performance. In order to evaluate the proposed method and compare its performance with other mainstream algorithms, we collect a continuous frame image dataset in a real-world environment for optical cooperative localization. Evaluation results based on location error under one-pass evaluation (OPE) show that the proposed method achieves seamlessness, performs optimally among similar methods, and is able to meet the real-time operation requirements at high resolution.