Performance Tests of a Proof-of-Concept Work Zone Intrusion Alert System in North Carolina Detected Intrusions Correctly 92 to 99 Percent of the Time.

METHODOLOGY

The study first analyzed work zone crash data in North Carolina between August 1, 2008, and January 31, 2020, to understand the crash types that occur due to work zones. Then two proof-of-concept systems were developed: a work zone intrusion warning system and a vehicle queue detection system. The work zone intrusion alert system was designed to alert workers through sounds and vibrations on their mobile devices. The queue detection system was designed to detect stopped vehicles inside a designated polygon work zone area. Both systems used AI-based object detection methods and were tested with publicly available highway traffic videos as the simulated test environment. 

FINDINGS

• Performance tests for the work zone intrusion alert system detected intrusions correctly 92.3 to 99 percent of the time. Detection accuracy was above 97 percent for both day and night during clear weather (see Table 1).
• Tests for the queue detection system had an accuracy rate of 93.4 percent during clear, daytime conditions. Nighttime and rainy tests were less successful, with rates of 80.7 percent and 66.6 percent accuracy, respectively (see Table 2). 

Table 1: Tests of the Work Zone Intrusion Alert System

Table 2: Tests of the Queue Detection System