Predicting Traffic Intensity with Deep Learning and Semantic Segmentation

• Activity Number: 256 - Contributed Poster Presentations: Section on Statistical Learning and Data Science
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Logan Bradley-Trietsch* and Xiao Wang
• Companies: Purdue University and Purdue University
• Keywords: Machine Learning; Deep Learning; Semantic Segmentation; Image Analysis; Convolutional Neural Networks; Traffic Prediction

Abstract

The computer vision task semantic segmentation, which involves assigning a label to each pixel in an image, has many applications to vital topics such as medical imaging or autonomous driving. This research project uses the semantic segmentation deep learning architecture Deeplabv3+ to make inferences about the intensity of vehicular traffic based on the vehicular pixel to total pixel ratio. This unique approach to traffic estimation has advantages over traditional methods like object detection in its ability to be robust against large vehicles. This project and its methods are one step toward the “smart city”, i.e., a city that has a traffic system based on artificial intelligence that can adapt to traffic in real-time with the end goal of mitigating gridlock traffic in large cities.