Date of Award
Master of Science (MS)
Civil & Environmental Engineering
A new approach for testing incident detection algorithms has been developed and is presented in this thesis. Two new algorithms were developed and tested taking California #7, which is the most widely used algorithm to date, and SVM (Support Vector Machine), which is considered one of the best performing classifiers, as the baseline for comparisons. Algorithm #B in this study uses data from Vehicle Re-Identification whereas the other three algorithms (California #7, SVM and Algorithm #A) use data from a double loop detector for detection of an incident. A microscopic traffic simulator is used for modeling three types of incident scenarios and generating the input data. Two incident scenarios are generated by closing either one lane or two lanes of a four-lane highway. The third scenario involves bottleneck blocking two lanes of the freeway with an incident occurring in the upstream of the bottleneck. The highway network is five miles long and simulated in VISSIM. Traffic parameters like occupancy, speed, flow and number of vehicles passing through the loop detector are collected to assess the traffic condition between the sensors or detectors. The proposed performance test inspects whether the algorithms thus tested were able to detect any occurrences and incidences within the first minutes in different scenarios and compares their respective detection to identify the best performing algorithm in all the contingencies. The results indicate that the implementation of this new approach not only reduces the dilemma of selecting thresholds but also checks algorithm performance in different incident scenarios so that the response time for clearing such incidences is as short as possible. Likewise, making use of Re identification data and travel time makes the incident detection more trivial and self-evident and thus outperformed the algorithms using traditional data like occupancy speed and volume in uncontested traffic conditions. Further different SVM models were trained and tested inspecting the effects of change in location of incident concerning detectors. However, using data from loop detector performed well when the incident happened at the upstream detector while using that from re-identification encountered delays in overall detection time for the same.
"Developing Algorithms to Detect Incidents on Freeways From Loop Detector and Vehicle Re-Identification Data"
(2019). Master of Science (MS), Thesis, Civil & Environmental Engineering, Old Dominion University, DOI: 10.25777/jt8g-yr58