Automatic traffic sign detection and recognition using moment invariants and support vector machine

Automatic traffic sign detection and recognition (TSDR) is one of the most significant areas of object detection. In spite of numerous researches, it has always been a challenging problem. In this paper, an approach for detecting circular and triangular traffic signs is proposed. The performance of the entire system is measured on German traffic sign detection benchmark (GTSDB) and German traffic sign recognition benchmark (GTSRB) dataset. Traffic signs are detected using color segmentation and thresholding method in Hue Saturation Intensity (HSI) color space. Then, the shape of traffic signs is detected using geometric invariant Hu moments. Further, the features are extracted using a technique called HSI-HOG descriptor where features are extracted from each channel of HSI independently. To select the most discriminant features with minimal loss of information, dimensionality reduction technique Principal Component Analysis (PCA) is applied and classification is performed using Support Vector Machine (SVM) technique.