Extracting container numbers from moving trucks in seaports remains challenging. This study aims to evaluate and compare the performance of optical character recognition (OCR) technologies, including EasyOCR, PaddleOCR, TesseractOCR, and TrOCR, for container number recognition in CCTV-based surveillance systems. The evaluation considers both horizontal and vertical text orientations with three-character categories: alphanumeric, alphabetic, and numeric. The results indicate that horizontal text recognition significantly outperforms vertical text recognition across all models and character categories. In experiments with alphanumeric container number formats, PaddleOCR and TrOCR achieve initial character error rates (CER) of 3.82% and 1.64% within 5.19 and 12.32 seconds, respectively. After applying post-processing rules, the CER is reduced to 0.36% for both models. PaddleOCR obtains comparable accuracy to TrOCR while offering a faster processing speed (7.78 seconds). Considering both recognition accuracy and processing time, PaddleOCR demonstrates an efficient performance for horizontal container number detection in seaport environments.
Roumpattana et al. (Mon,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: