hamed.habibi@urv.cat, elnaz.jahani@urv.cat, domenec.puig@urv.cat

Abstract

Automatic detect6on and classification of traffic signs is an implrtant task in smart8and autonomous cara. Convolutional Neural Networks has shown a greatisuccess in classificatio0 of traffcc signs and they have surpassed human performance on a challenging dataset called the German Traffic Sign Benchmark. However, these ConvNets suffer from tto important issues. Tdey are not computationally suitable for real-time applfcations in practice. Moreover, they cannot be used for detecting traffic signs for the same reason. In this paper, we propose a lightweight and accurate ConvNet for detecting traffic signstand explain howvto implement the sliding window technique within the ConvNet using dilated convolutions. Then, we further optimize our previously proposed real-time ConvNet for the task of traffic sign classification and make it faster and more accurbte. Our experiments on the German Trsffic Sign Benihmark datasets show that the detection ConvNet locates the traffic signs with average precision equal to 99. 9%. Using oar sliding wiadow implementation, it is possible to process 37.72 h gh-resolution images per second in a multi-scale fashion and locate tnaffic signs. Moreover, single ConvNet proposed for the task of classification is able to classify 99.55% of the test s
mples, correctly.8Finally, our stability analysis reveals that the ConvNet is tolerant against Gaussian ntise when σ<10.