Improving the Classification Accuracy of Accurate Traffic Sign Detection and Recognition System Using HOG and LBP Features and PCA-Based Dimension Reduction

Abstract

The automatic traffic sign detection and recognition (TSDR) is one of the most useful, important and consequential method towards designing automated and driverless vehicles. It is one of the real challenges to overcome. Even in current vehicles with drivers, it is an important matter which need to be tackled for enhancing driver’s safety. In this research, we propose to use histogram-oriented gradient (HOG) and local binary patterns (LBP) approaches for feature extraction followed by PCA-based dimension reduction for accurate TSDR system. The full approach and performance are measured on Chinese traffic sign (TS) database. The development approach of the system is divided into three steps: Image processing, detection, and recognition. For the detection of traffic signs, we used an RGB color segmentation (thresholding) technique with circular Hough Transform for circular TSDR and Hough transform Algorithm for shape detection. For recognition, features are extracted using techniques HOG and LBP. Further principal component analysis (PCA) is applied for strong component extraction and classification is performed using support vector machines (SVM) classifier. The result section validates the effectiveness of using the proposed features, as the classification accuracy has been improved with respect to previously reported works on TSDR.