Aiming at developing underwater battery recharging system, we have been researching on automatic docking of an underwater robot using stereo-vision-based visual servoing and 3D marker. The docking function deems to be an important role not only for battery recharging but also for other advanced applications, such as information transmissions. The authors have proposed a optical docking system and conducted real sea experiments to verifythe practicabilityof the proposed system composed of stereo-vision-based 3D pose (position and orientation) realtime measurement system. However, our proposed system sometimes failed the docking operation in the dusk and turbid environment since our recognition method lost the 3D marker in the natural lighting environment that changes every moment. In this paper, therefore, we proposed a new fitness function for improving the robustness against the lighting change. To improve the robustness, firstly, we propose a fitness function composed of color (HSV) and brightness evaluation for overcoming the difficulties to estimate in realtime 3D pose of the underwater vehicle in lighting and turbid varieties. Secondly, we modify the fitness function to improve sensitivity by adding a heuristic rule that increases the evaluation value when the all color balls of the model overlap real 3D marker in the camera images. This approach enables our docking system to apply to the natural lighting environment that changes every moment for increasing success rate of the docking operations. Thirdly, the effectiveness of the proposed fitness function adaptive to the changing lighting environment has been confirmed in the outdoor pool environment. Finally, our proposed docking system has been verified to be robust against lighting environment varieties, by successful repeated docking experiments in turbid environment with lighting condition changes from daytime to sunset in real sea.