Abstract
This paper presents a new method for estimating traffic density on freeways, and an adaptation for real-time applications. This method uses re-identified vehicles and their travel times estimated from a real-time vehicle re-identification (REID) system which attempts to anonymously match vehicles based on their inductive signatures. The accuracy of the section- 6 based density estimation algorithm is validated against ground-truth data obtained from recorded video for a six-lane, 0.66-mile freeway segment of I-405N in Irvine, California, during the morning peak period. The proposed density estimation algorithm results are compared against a g-factor based method which relies on inductive loop detector occupancy data and estimated vehicle lengths from the Caltrans Performance Measurement System (PeMS) as well as a selected REID method which uses a sparse REID algorithm based on long vehicle detection and volume counts at detector stations. Although the g-factor approach produces real-time density estimates, it requires seasonally calibrated parameters. In addition to the calibration effort to maintain overall accuracy of the system, the g-factor approach will also produce errors in density estimation if the actual composition of vehicles yields a different observed g-factor from the calibrated value. In contrast, the proposed method uses an existing vehicle re-identification model based on the matching of inductive vehicle signatures between two locations spanning a freeway section. This approach does not require assumptions on the vehicle composition, hence does not require calibration. The proposed algorithm obtained section-based density measures with a mean absolute percentage error (MAPE) of less than four percent when compared against groundtruth data and provides accurate density estimates even during congested conditions, improving both the PeMS and selected alternative REID based methods.
