Automatic Container Code Localization and Recognition via an Efficient Code Detector and Sequence Recognition

The challenge of container code localization and recognition is that the code varies in size, position, and arrangements according to the container style and environmental changes. This paper proposes an automatic container code localization and recognition system via an efficient code detector and a sequence recognizer. The code detector locates text features on the backend of the container and extracts the code through segmentation, which increases the stability of code localization. The proposed recognition algorithm regards the code as three separate sequences and recognizes them through a recurrentconvolutional neural network to improve efficiency and accuracy. To verify the validity, a test dataset is established containing about 700 high-resolution images and over 2000 gray-scale code patches. The proposed code localization and recognition algorithm is verified on the dataset in experiments and achieves overall accuracy of 93.98% at about 0.1s per-frame.