Traffic sign recognition using visual attribute learning and convolutional neural network

The problem of extracting high level information from digital images and videos is frequently faced in the area of computer vision and machine learning. For the recognition of traffic signs, a lot of outstanding methods have been proposed, and deep models demonstrates that their powerful representation capacity, can archieve dominant performances. In this paper a method for recognizing traffic signs is proposed founded on a novel visual attribute mechanisms; whereby attributes are generated using Convolutional Neural Networks (CNN). In comparison with previous methods founded on the use of CNN for feature extractor and Multi-Layer Perception (MLP) as classifier, the Max Pooling Positions (MPPs) proposed in this paper predict visual attributes that provide a useful linkage between low-level features and high-level sematic tasks. The results show that outstanding performances can be achieved using MPPs.