Detection and recognition of road signs using simple layered neural networks

A road sign detection method is proposed, using 2 simple 3-layered neural networks. Multiple preprocessing steps are taken for masking the irrelevant areas and selecting the candidate areas from an original input color images, and both neural networks are modules for this selection process. One network is for matching a color of an input pixel with a road sign, and the other is for matching a shape of an input object. The final recognition is given by a template matching, and an auxiliary re-detection step is added to improve the efficiency. The experiments using a large number of pictured images under several different conditions show the high detection rates over 95% in most cases, while the computational cost is low owing to the smallness and the simplicity of the neural networks.