earticle

영어

The time information recorded in a vehicle black box or CCTV is displayed in the text time(ex: YYYY/MM/DD hh:mm:ss). However, when restoring images from inactive areas, temporary files, or damaged video files, there could be a problem that the sequence of images restored to frame units is not arranged in that of the actual recorded time. To resolve this issue, we recognize time information recorded in text format at the top or bottom of the image, and rename the image file via the recognized time information. At that time, images that have been renamed and sorted chronologically can be made into continuous videos. In this paper, Optical Character Recognition(OCR) was used to recognize time information. OCR-related research has continued for a long time and has recently been available through many open sources. Thus, time information of the image is extracted by applying Convolutional Recurrent Neural Network(CRNN) based OCR, Tesseract OCR, and EasyOCR. The expression of time information in vehicle black boxes and CCTV is expressed in various ways(ex: text font, background and text size). The performance of each OCR was compared and analyzed through test images of time information presented in various forms. It was shown from the comparative analysis that the CRNN-based method has the highest recognition performance among OCRs used in the experiment. Moreover, it was confirmed that a prediction rate is approximately 5 times faster for predicting a single image.