Forbidden Traffic Signs Detection and Recognition Based on Sparse Representation

This paper presents an automatic traffic-sign detection and recognition algorithm based on sparse representation. Our system consists of several steps. In the first stage, we detect potential traffic signs using the most remarkable feature-color-extracted by HSI model, which is immune to lighting changes. In the second stage, the preprocessing stage, image binaryzation and image cutting help the system to extract candidate traffic-sign regions. Then OMP algorithm is performed to calculate the sparse coefficients of candidate traffic-sign regions on dictionary D. The dictionary D is constructed by training images with different rotation. Finally, the sparse coefficients are used to classify. The algorithm proposed offers high performance and better accuracy especially in variable lighting conditions and rotation. Because of its real time and accuracy, this algorithm can be used in real world application.