Analysis of Various Traffic Sign Detectors Based on Deep Convolution Network

A lot of progress has been made on the object detection field, and it is difficult for researchers to decide which architecture is suitable to their specific task. The work in this paper aims to compare and analyze various detection models and then obtain an appropriate model for the traffic sign detection task. Two main components of a detection model are meta-architecture and feature extractor, and we will explain these two parts on the paper. Experiments show that the behavior of the detection model is strongly affected by the choice of architecture and feature extractor. Also, Region-based Fully Convolutional Networks (R-FCN) architecture can achieve better performance than Faster Regions with Convolutional Neural Network (Faster R-CNN) and Single Shot Detection Network (SSD) in traffic sign detection task.