In this work, we read licence plates using Optical Character Recognition. Our algorithm relies on the detection of plates on the basis of contours and text that they contain. Thank to the combinaison of these two detection mode, our algorithm remains effective even when the plate is slightly obstructed. For the separation of false positive from real images of license plates, we use filters based on text and registration plates general characteristics. This gives our algorithm a great ability to adapt to different contexts. Following our experimental test on license plates in Benin, we obtained a recall rate of 86% and an accuracy rate of 60%.
展开▼
