Computer Vision on Vehicular Flux-Density Relationship Regression
The classical Greenshield's model for one-dimensional traffic postulates a linear relationship between a parameterized automobile density \(\rho\) and velocity \(u\): \(u = 1 - \rho\). Under the guidance of Anasse Bari, Xinran (Kevin) Li and I found that the empirical assumption could be further refined by utilizing the novel object detection algorithm YOLOv8, and therefore regressed for a more precise density-velocity relationship for any specific sufficiently-long one-dimension roadway. With a case study of I-95/495 on Woodrow Wilson Bridge in Maryland, we attempted to outline an object detection approach of refining Greenshield's model with dynamically adaptive parameters subject to local data. This refined one-dimensional traffic model could be particularly favorable in certain roadway conditions - tunnels, elevated bridges - so as to ease traffic congestion in accordance with model suggestions.
See here for our poster presentation that won the best presentation award at the 2024 NYU Undergraduate Research Conference.
