AN EFFICIENT ALGORITHM FOR IMPLEMENTING TRAFFIC SIGN DETECTION ON LOW COST EMBEDDED SYSTEM

摘要

This paper presents the implementation of traffic sign detection on a low cost embedded system. The proposed system consists of color thresholding, shape detection, and sign validation. An efficient color thresholding based on the red-blue angle color transformation (RBAT) and the normalized red color is employed to extract the red circular traffic sign in an image. Then an ellipse fitting technique is applied to detecting the circular sign. Further, a novel technique to validate the circular sign is proposed by employing the histogram of oriented gradient (HOG) which is calculated on the outer area of the sign. The experimental results show that the true positive rate of 0.92 and the precision of 0.97 could be achieved. The execution time of the proposed algorithm which is implemented on a Raspberry Pi Type 3 module is 48.17 ms. It provides a promising result for the real-time implementation.