One-Shot Traffic Analytics

点击翻译按钮获取中文摘要

In this project, we investigate the possibility of monitoring highway traffic as well as street intersection based on a single image, i.e. without using motion features. The goal of the proposed method is to recognize traffic conditions instead of measuring it as is usually the case. The main objective of this project is to allow an automatic system to process low frame-rate videos for which motion features cannot be estimated. The method that shall be developed takes advantage of the highly redundant nature of traffic scenes which are pictured from a top-down perspective and made of vehicles pictured on a predominant asphalted road surrounded by background objects. Due to the limited variety of objects pictured in traffic scenes, our method shall get to learn features that are unique to such images. With these features, the method will look forward at segmenting traffic images, classify traffic scenes, and estimate traffic density without the need for accurate motion features. We consider using a convolutional neural network (CNN) model that we recently developed. From that model, different CNN will be proposed to segment traffic images in three different classes (road, car and background), classify traffic images into different categories (empty, fluid, heavy, jam) and predict traffic density.