Hierarchical Traffic Sign Recognition Based on Multi-feature and Multi-classifier Fusion
Yunxiang Ma, Linlin Huang · Advances in computer science research · 2015
In this paper, we propose a fast and robust method for traffic sign recognition, which uses a coarse-to-fine strategy.The traffic signs are divided into main category and sub-category.At the coarse classification stage, we extract histogram of oriented gradients (HOG) feature from different spectral bands of traffic sign images and classify into main category using a linear support vector machine (SVM).Then at the fine classification stage, complementary features of dense-sift, local binary pattern (LBP) and Gabor filter features are extracted, fused and then fed to a committee of SVM and random forest.The proposed method gets an accuracy of 98.76% on the German Traffic Sign Recognition Benchmark (GTSRB) dataset and takes about 50ms per image.Both recognition accuracy and speed is higher than that of the method based on multi-scale convolutional neural network.