A Study on Traffic Sign Detection and Classification with Single Shot Detection

摘要

In this paper, we would be presenting our basic investigation on real life traffic-sign detection and classification in the form of images. We applied one of the latest state of art in object detection on the public dataset that was submitted to Conference on Computer Vision and Pattern Recognition (CVPR) in 2016. The dataset we would be working with was submitted as part of the paper "Traffic-Sign Detection and Classification in the Wild". Unlike various dataset that was available in the past, this traffic sign dataset could represent the images encountered in real life. For their own experiment, they used R-CNN based network to detect and classify the traffic sign, which is not the latest and fastest method that is currently available. Thus, we decided to conduct some experiments with the dataset using different method, Single Shot MultiBox Detector, which is one of the most accurate and fastest methods that are currently available. We will describe our observation on several experiment conducted in order to optimize the detection speed and the accuracy for this particular dataset.