Video Analytics Usage in Transportation Agencies: Nationwide Survey and Maryland Feasibility Study

Abstract

This paper summarizes the findings from a study conducted by the University of Maryland Center for Advanced Transportation Technology (CATT) regarding the potential to extract meaningful traffic information from existing Maryland Department of Transportation-State Highway Administration (MDOT-SHA) closed-circuit television (CCTV) video feeds via image processing. It describes the survey that was developed and administered by CATT between March 10 and June 8, 2017, focusing on transportation agencies’ current use cases and future plans for video analytics. Feedback from 21 agencies (9 of which are using video analytics) indicates a general excitement about the technology, although a persistent theme throughout the responses is that most current solutions are not yet able to provide satisfactory quantitative results (e.g., traffic counts, speeds, turning movements), especially in low light and high glare scenarios. Additionally, the report summarizes the vendor evaluation procedure that was undertaken, which includes identifying suitable video analytics vendors, developing a testbed of representative video clips from MDOT-SHA cameras under various conditions, asking the vendors to demonstrate their product capabilities on the testbed, and analyzing results. With proper camera positioning and calibration, the two vendors who participated achieved results within 5% to 15% of manual counts, and 4% to 7% of probe speeds in the northbound travel direction during a one-hour test clip. However, only one vendor produced results for the more challenging southbound direction, and the estimates were far less accurate (within 25% of manual counts and 20% of probe speeds). Accordingly, it appears that the estimation accuracy is highly sensitive to factors such as camera angle, resolution, and visibility.