An automatic bus route number recognition

Abstract

This paper presents a bus route number recognition using image processing and machine learning techniques. The prototype is designed to extract bus route number from scene images and synthesize audio of bus number for people with low vision. The proposed system consists of two parts, text detection and digit recognition. Text detection based on MSERs is used to find and segment text area from input images. Digit recognizer based on Convolutional Neural Network is used to recognize digits on the detected text area. Finally, the recognized digit is converted to audio which can help the visually impaired people (low vision) to take the correct bus. The experimental results of the proposed method presented in precision, recall and F-measure in text detection and acquire 73.47% accuracy of a digit segment in self-collect Thai bus dataset.