Usage of image recognition and classification with Artificial Intelligence (AI) is prevailing in a wide range of sciences.  Long-term visual and ecological data, e.g., still photographs and videos, sometimes do not get on with those advanced techniques.  Obstacles are insufficient resolution, noises caused by weather conditions, etc. Classical methods for image analyses would be robust and instead preferred in such cases. We applied a simple opening/closing technique for detecting herons and egrets in three-year (2018-2020) video footage from 80m above a gigantic colony at the Kashima industrial complex. The video was taken with a network camera attached at the top of the observation tower for a windmill. Not all individuals in the colony were breeding, and most of them were diurnal. So we recorded the colony at the break of dawn for grasping the whole constituent members. By calibrating with manually counted reference data obtained with a UAV in 2015, we reconstructed three-year dynamics of the colony in which thousands of herons and egrets gathered for breeding. It might be the first to report such a vast colony's population dynamics for multiple years with automatic recording.