Speed Limit Sign Recognition Using MSER and Artificial Neural Networks

An efficient real-time speed limit sign recognition system could provide significant benefits for realizing advanced driver assistance systems (ADAS). This paper presents an approach for real-time recognition of U.S. speed limit sign. Detection of speed limit sign is carried out using shape and intensity information after identifying the candidate regions as maximally stable extremal regions (MSERs). The detected sign is tracked through the subsequent image sequences using Kalman filter. Finally, artificial neural networks based classifier is used for the recognition of speed limit sign. About 98% correct recognition with an average processing speed of about 40 fps on a standard PC is achieved for 12300 images of different conditions.