A Tool for Images Validation in Cell Sites using Text Recognition in Natural Scenes

Abstract

In the construction, installation, and maintenance of Radio station, employees need to create reports with information and real photos to prove that each provided service was accomplished. The creation of this report is generally slow, costly, and unpredictable. This occurs mainly due to the manual process involved in the incorrect acquiring of the images. On the other hand, computer vision techniques can significantly decrease the time and cost of this activity, avoiding illegible or incorrectly captures of the station board images. Thus, this work aims to propose a mobile tool to perform a validation of these images of the station board, using computer vision and artificial intelligence techniques. Thus, a tool was developed using the Python language, the pre-trained network EAST, and the Tesseract and Kivy libraries. We validated the approach in real-world cases, and the method was able to extract the default key text correctly. However, in non-board images, the proposal still needs some tuning to extract the key-text correctly. The primary goals of the research were accomplished since the tool was able to perform a validation of the board image. We intend to include new strategies to improve text recognition capabilities.