Visual Recognition of Container Number with Arbitrary Orientations Based on Deep Convolutional Neural Network
Wanghui Bu, Shuang Zhi Yan, Jing Chen, Chenguang Yang, Chengju Liu · 2018
The visual recognition of container number is vital for unmanned ports. Most of the existing methods focus on the simple case where the orientation of the container number is in horizontal direction. However, the orientation of the container number in images captured by the camera in unmanned ports is often inclined. Hence, this paper presents an approach to complete a pipeline for visual recognition of container number with arbitrary orientations based on deep convolutional neural network (CNN). In order to deal with various complex cases, the processing of the pipeline is divided into three stages, namely the coarse localization, the fine localization and the text recognition. The proposed approach is implemented by TensorFlow. Experimental results show that the pipeline realizes excellent detection and recognition of container number with arbitrary orientations and various arrangements. According to the experimental results, the total accuracy of the proposed approach is over 85% at present.