Deep Learning Driven Traffic Sign Board Detection for Real Time Road Safety Monitoring

Abstract

Traffic signs are road facilities that communicate, direct, limit, caution or teach information, whether in the form of words or symbols. As the demand for the intelligence of vehicles is on the rise, there is a great need to invent and identify traffic signs automatically using technology. Nonetheless, the identification of traffic signs is not that easy, as a number of negative parameters exist, such as bad weather, change of perspective, physical impairment, and others. Currently, most of the available text mining algorithms help in processing the whole data to identify the traffic sign images. In this proposed research, an extensive sign board detection algorithm is developed where AlexNet image classification algorithm forms the premier stage. It is mainly focussed on the process of detection with the improvement of the traffic signs using a boundary enhancement algorithm along with the average filter. This helps in reducing the noise and enhances the sign to be fed into the classifier system. This approach enhances precision of 99.27%, sensitivity of 99.41% and specificity of 99.47%. Thus the proposed algorithm minimizes the time taken to detect the traffic sign in misty weather.