Multi-column spatial transformer convolution neural network for traffic sign recognition

Abstract

Traffic sign recognition is an important research area in intelligent transportation, which is especially important in autopilot system. Convolutional Neural Network (CNN) is the main research method of traffic sign recognition. However, the convolution neural network is easily affected by the spatial diversity of the image. With regard to this, in this paper, a multi-column spatial transformer convolution neural network named MC-STCNN is proposed to solve the problem when Convolutional Neural Network (CNN) can’t adapt to the spatial diversity of the image very well. The MC-STCNN network consisted of CNN and STN is formed by training pictures of different sizes. It can be well adapted to the spatial diversity and the images input of different sizes. It achieves an accuracy of 99.75% on GTSRB traffic sign recognition, exceeding the current highest accuracy of 99.65%.