An Effective Method for Automated Railcar Number Detection and Recognition Based on Deep Learning
Ran Zhang, Zhila Bahrami, Zheng Liu · 2021
Railway transportation is of great value to the economy. Railcars usually travel at fast speeds in complex environments. Accurately detecting and recognizing the railcar numbers contributes significantly to logistic management. Therefore, this paper presents a deep-learning-based method for automated railcar number detection and recognition. It consists of two steps, i.e., railcar number detection and recognition, which are conducted by two separated deep learning models. In the detection process, both the whole region of railcar number and text instances are detected so our method is able to remove the detected noisy text regions, i.e., false positives. The cropped text areas in the railcar number regions are sent to the text recognizer. Crucially, the text is recognized on word level so there is no need to conduct an extra step to separate all characters. Experimental results on the railcar number dataset demonstrate that our proposed method can effectively detect and recognize railcar numbers.