Project: #318 Determining segment and network traffic volumes from video data obtained from transit buses in regular service: Developments and evaluation of approaches for ongoing use across urban networks Progress Report - Reporting Period Ending: Sept. 30, 2022 Principal Investigator: Mark McCord Status: Active Start Date: June 1, 2020 End Date: June 30, 2023 Research Type: Advanced Grant Type: Research Grant Program: FAST Act - Mobility National (2016 - 2022) Grant Cycle: Mobility21 - The Ohio State University Progress Report (Last Updated: Sept. 30, 2022, 8:44 a.m.) % Project Completed to Date: 95 % Grant Award Expended: 90 % Match Expended & Document: 97 USDOT Requirements Accomplishments Goals and Objectives • Refine approaches to determine traffic volumes and related network measures in urban areas from video sensors mounted on transit buses in operational service • Demonstrate performance of the method • Motivate practical use of the method and begin applications • Expand method to an approach for ongoing monitoring of traffic measures Accomplishments • Three new undergraduate students were recruited and trained in vehicle digitization by an existing undergraduate student. • Vehicle miles travelled (VMT) were estimated and compared on subsets of networks of roadways on the campus of The Ohio State University (OSU) on days in multiple years using segment traffic volumes determined from video data, estimates obtained from a popular on-demand location-based-service (LBS) data integrator and provider, and road-tube data. VMT was estimated for a 10-hour period (8:00 am to 6:00 pm) and for each hour in this period to investigate time-of-day patterns. Comparisons were made to road-tube based VMT as “ground truth” on networks where the road-tube data were available and to patterns based on knowledge of traffic patterns across years. Results showed very good correspondence to ground-truth and known patterns when using video-based traffic volumes and very poor correspondence when using LBS-data-based traffic volumes. • Bus-based video imagery was collected on a weekly basis for one segment-direction for two different hours on different days of the week while traffic volumes were manually observed for the same segment-direction-hours to serve as ground truth. Much of the video-based imagery was processed to estimate segment-direction-hour traffic volumes, which are being compared to the manually collected ground truth to investigate accuracy in estimating an average-day volume from bus-based video imagery over multiple days. As expected, empirical results show that differences between video-based and ground-truth volumes estimates decrease appreciably when estimating traffic volumes on an “average day” compared to an individual day. The ability to capture day-to-day and hourly variations in traffic volumes from bus-based imagery was also begun. • Regression-based transformations of video-based volumes estimated using present volume estimation techniques were investigated. The transformations are based on observable characteristics, namely, the untransformed video-based volume estimate and the equivalent observation duration associated with the video imagery. However, equivalent observation duration were not found to impact the performance of the regression model-based transformations. Regression-based transformations were estimated using data on a calibration sample of road-tube and video-based estimates from some years and roadway segments and applied to a holdout sample for other years and segments. The transformations were seen to work well when estimated video-based volumes were relatively high, but poorly when volumes were relatively low. • Regression-based models to predict the error in video-based estimations as a function of estimated traffic volumes and equivalent observation durations were estimated. The objective was to use these error models to indicate confidence in a specific video-based estimate either for qualitative or quantitative (e.g., weighting of estimates when determining averages) purposes. Although the estimated models showed statistical significance associated with explanatory variables, validation results are not indicating useful predictive power. Training and professional development • Three graduate students and eight undergraduate students were involved with various efforts of this project. • One undergraduate student continued to take responsibility for training students on various video imagery processing tasks and supervising imagery processing assignments. • A second undergraduate student took a leadership role in designing and implementing the logistics of weekly manual traffic observations on a select roadway segment. She also started analyzing the ability to determine daily variations in traffic volumes obtained from video imagery during this reporting period using the collected data. The student graduated from the undergraduate program and began graduate studies, where she continues her involvement with this aspect. Dissemination • A paper was submitted for possible presentation at the January 2023 meeting of the Transportation Research Board and for possible publication in Transportation Research Record. This paper demonstrates the good accuracy in estimating VMT and time-of-day patterns in traffic volumes across networks when using video-based traffic volume estimated with methods developed in this project and the much better accuracy in these estimates compared to commercially available estimates determined from Location-Based Services data. • Investigators meet regularly with administrators from The Ohio State University’s Transportation and Traffic Management (TTM) to update them on this and other projects that address practical issues of interest to TTM operations. Upon request of the investigators, TTM provides the video imagery used for the research, outreach, and educational tasks in this project. Plans for upcoming period • Process additional video imagery into digital location- and time-specific vehicle observations • Determine volume estimates from additional processed video imagery • Validate the benefits of estimating average-day traffic volumes in a time-of-day period relative to estimating single day traffic volumes in the time-of-day period • Investigate the ability to determine daily and hourly variations in traffic volumes from bus-based video for use in ongoing traffic monitoring • Continue interactions with OSU Transportation and Traffic Management (TTM) Impacts • Regular meetings with The Ohio State University (OSU) Transportation and Traffic Management (TTM) office continue to generate interest in the empirical traffic flow estimates being produced and motivate ongoing collaboration in project research and outreach efforts Other Physical collections: During the period video imagery and manually collected data were obtained Curricula: During the period, the approach being refined and evaluated in this study was presented as a module in one undergraduate/graduate Civil Engineering class consisting of 28 students. Outcomes New Partners Investigators continued to work closely with Transportation and Traffic Management (TTM) at The Ohio State University (OSU). TTM oversees all OSU transportation operations, other than parking, and operates the Campus Area Bus Service (CABS), with a fleet of approximately fifty 40-foot transit buses serving close to 5 million passengers per year on fixed route scheduled services. Issues None