In this work, we address the issue of traffic sign detection that can scale to several hundred classes of traffics signs and can be efficiently applied to traffic signs with varying sizes and aspect ratios. We propose to employ state-of-the-art region proposals as the first step to reduce the initial search space and provide a way to use a strong classifier for a fine-grained classification. We evaluate multiple region proposals on the domain of the traffic sign detection and further propose various domain-specific adaptations to improve their performance. We show that Edgeboxes with domain-specific learning and re-scoring based on trained shape information are able to significantly outperform remaining methods on German Traffic Sign Database. Furthermore, we show they achieve a higher rate of recall with high-quality regions at the lower number of regions than the remaining methods. Although the proposed approach was evaluated on the domain of traffic signs, the developed methods are independent of any specific domain and can be applied to any other domain, such as detection of cars or pedestrians.
In this paper, we present our initial research on traffic sign recognition, done in the context of developing an automatic system for traffic sign inventory. We revisit the problem of the traffic sign classification, and investigate the performance of linear support vector machines and extreme learning machines, which are often sidelined in favor of more complex, high-performing approaches. We perform extensive evaluations on two standard traffic sign classification datasets, showing that these relatively unsophisticated classifiers are nevertheless capable of achieving competitive performance. Furthermore, these approaches are readily extensible to incremental learning, and therefore present a lucrative choice for applications in traffic signalization detection and recognition.