The surge in global trade necessitates enhanced efficiency in logistics operations, particularly in container management. Current manual identification methods for container license plates and labels are prone to errors, causing significant delays and increased emissions. This paper presents a divide-and-conquer approach for detecting container license plates, distinguishing between the identification of numbers and letters, and leveraging multiple video frames to improve detection performance. We designed a set of experiments to define which models are optimized for each task, namely we assessed the performance of models to identify the whole license plate and models specifically for numbers and letters. Results on a real dataset of shipping containers entering dry ports show that the divide-and-conquer approach surpasses the holistic approach. Additionally, experiments show that for different contrasting colors analyzing multiple frames can improve the identification performance.