Project Proposal for Traffic reconstruction
Susu Li and Lisa Lyons {susu, llyons}@unc.edu University of North Carolina at Chapel Hill Department of Computer Science
1. Introduction
This is a project proposal for the Robotics course final project involving reconstructing traffic flows to create a realistic visualization of highway and streetlevel traffic for synthetic environments. This project will integrate a new model – traffic lights - into a project that has been underway for some time in the Gamma Group, namely Traffic Simulation and Reconstruction.
2. Previous Work
Jur Van den Berg and Jason Sewall in [1] have implemented a novel method of Virtualized Traffic in which traffic needs to be reconstructed from discrete data obtained by sensors placed alongside a highway or street. They have presented an algorithm to determine the trajectories for multiple cars that also allows streaming realworld traffic data in real time to visualize traffic as the data comes in.
3. Problem Justification
The problem is defined as follows: We construct a roadmap in an urban environment which contains multiple roads, intersections and traffic lights, as well as several sensors. Given the data generated by two sensors at point A and B, we reconstruct a continuous stream of cars of the roadmap between these two points of length L that has N lanes of a certain width. At sensor A, the data will be ( , , ), i є 1, 2…N; where є R is the time at which car i passes the point A, є is the velocity of car i when it passes the point A, є 1, 2…N is the lane in which car i passes the point A. At sensor B, the data will be ( , , ), i є 1, 2…N; where є R is the time at which car j passes the point B, є is the velocity of car j when it passes the point B, є 1, 2…N is the lane in which car j passes the point B. At the traffic light , the data will be = 1,
є 0, 30 60,90 … … 0,
for
example. When L =1, the traffic light turns red, otherwise it turns green.
The task is to compute trajectories for the cars on the road which satisfy the following constraints: 1) at the point A at the time , … …the cars go through with the velocities of , … …in the lane , … …; 2) The same for sensor B; 3) If there are traffic lights between points A and B, all cars will avoid going through when the lights turn red. 4) The cars should respect geometric constraints (Van den Berg and Jason Sewall have implemented this). 5) Each car should maintain sufficient distance from every other car and drive smoothly – without unnecessarily accelerating. 6) For the intersection with multiple lanes, cars can turn right left and U-turn under traffic rule with no collision.
4. Research Design
Since the project is based on Van den Berg and Jason’s previous work, we will start from the simplest case, and then extend it to more complicate cases: i. ii. iii. Only a traffic light on a road with one lane between points A and B; An intersection with two single-lane roads between points A and B; An intersection with multiple lanes roads between points A and B(implement the left turn light, right turn light and U-turn here); iv. Multiple intersections in between points A and B;
5. Time line
Oct. 10, 2008 Nov. 10, 2008 Dec. 04, 2008 - Project Proposal - Have single lane road with stoplight implemented (Progress Report) - Have multilane road with lights implemented (Final Project Presentation & Demo)
6. Novelty
The novelty of this project is obvious. It’s an extension based on Van den Berg and Jason’s previous project, and the new model - intersections with traffic lights or stop signs – can extend the roadmaps from highways to the urban/suburban environments.
7. Other possible improvement
1) Stop signs could be implemented as well as lights.
2) Then change the data in [1]-for every car, there are given time and velocity at each sensor - to there is no specific time and velocity for each car at first. 8. References [1] J. van den Berg, Jason Sewall, Ming Lin and Dinesh Manocha. Virtualized Traffic: Reconstructing Traffic Flows from Discrete Spatio-Temporal Data