Abstract:
Learning a good generative model is of utmost importance for the problems of computer vision, image classification and image processing. In particular, learning features from small tiny patches and perform further tasks, like traffic sign recognition, can be very useful. In this paper we propose to use Deep Belief Networks, based on Restricted Boltzmann Machines and a direct use of tiny images, to produce an efficient local sparse representation of the initial data in the feature space. Such a representation is assumed to be linearly separable and therefore a simple classifier, like softmax regression, is suitable to achieve accurate and fast real-time traffic sign recognition. However, to achieve localized features, data whitening or at least local normalization is a prerequisite for these approaches. The low computational cost and the accuracy of the model enable us to use the model on smart phones for accurately recognizing traffic signs and alerting drivers in real time. To our knowledge, this is the first attempt that tiny images feature extraction using deep architecture is a simpler alternative approach for traffic sign recognition that deserves to be considered and investigated.
