Real-Time Bus Recognition for Adaptive Signal Control

Source Organization:
University:Carnegie Mellon University
Principal Investigator:Stephen F. Smith, Xiao-Feng Xie
PI Contact Information:sfs@cs.cmu.edu412-268-8811
Project Manager:Stephen F. Smith
Funding Source(s) and Amounts Provided (by each agency or organization):UTC - $100,000
Total Dollars:$100,000
Agency ID/Contract/Grant Number:DTRT12GUTC11
Start and End Dates:01/2013-12/2013
Project Status:Complete
Subject Categories:Operations and traffic management
Abstract:The ability to detect buses in oncoming traffic in real-time offers unique opportunities to improve overall traffic flow in urban environments. Buses regularly disrupt traffic flow as they pickup and discharge passengers. Yet, if traffic flows at a given intersection are not simultaneously blocked in multiple directions, there are often traffic signal control decisions that can be taken adaptively to minimize these disruptive effects (e.g., by servicing cross traffic) and reduce overall traffic congestion. Existing adaptive traffic signal control systems do not attempt to recognize and act upon the presence of buses in incoming traffic streams. Alternatively, existing approaches to bus prioritization start from the assumption that bus movement trumps all other vehicles, give no attention to how disruptive it is to overall traffic flow to keep buses moving, and relies on additional hardware, both within the vehicles and at each intersection. We propose to investigate development of the capability to use video streams from commercial traffic cameras to detect the presence of buses in real-time and integrate the use this information into an adaptive traffic signal control scheme.
Describe Implementation of Research Outcomes (or why not implemented):
Impacts/Benefits of Implementation (actual, not anticipated):We have collected training samples from current video streams at our East Liberty pilot deployment and used these images to develop a set of bus recognition exemplars. Evaluation of a recognizer that utilizes these exemplars has shown near 100% accuracy for recognizing buses close to the intersection, with degradation as the distance to the intersection increases.

Current work is focused in two directions. First, we are investigating the potential of boosting longer-range recognition by exploiting temporal sequences of images. Second, we are analyzing ways to optimize the efficiency of the current recognizer to enable real-time operation. Our plans over the next 6 months are to field test the bus recognizer with the SURTRAC adaptive signal system at an appropriate intersection in East Liberty.
Project URL:surtrac brochure.pdf