MSER-based text detection and communication algorithm for autonomous vehicles

Text detection and communication in the automotive context has attracted the attention of researchers only over the past few years. Detecting text in the automotive context, as opposed to the detection of text of printed pages, imposes additional challenges, such as the presence of obstacles, blurry frames, speedy vehicles, etc. In this paper, we present an in-vehicle real-time system able to localize texts and communicate them to the drivers. Our system begins by localizing regions of interest as a Maximally Stable Extremal Regions (MSERs). Afterwards, we apply a novel filtering stage which begins by dividing each ROI into 4 × 4 cells, counting the number of edges in each cell and comparing them to a well defined threshold. This is performed in order to filter out a considerable number of unwanted objects. The ROIs that contain text are fed into a recognition module based on the Optical Character Recognizer (OCR). Our proposed method achieves high f-scores when tested against several videos containing a numerous panels.