Edge-AI Enabled Automated Flaggers for Roadway Work Zone Management

Work zones are common for preservation and enhancement of roadways. Several efforts have been made to both improve safety for workers, especially, the flaggers, and reduce operational costs of work zones. In this paper, we investigate and prototype low-cost edge-AI enabled automated flaggers for traffic control to eliminate the need or reduce the number of human flaggers. To this end, we build a pair of connected portable traffic light units, each equipped with a camera, built-in microprocessor, portable battery and long-range WiFi radio for deployment at two ends of a work zone. We then develop an auto-flagger system with two modes: one semi-automated and the other fully-automated, by leveraging state-of-the-art deep learning and computer vision techniques. The semi-auto flagger system is trained to monitor and recognize hand-signals of a single human-flagger who stands at a safe location within the work zone, which then controls the two smart traffic light units. In the fully-automated flagging mode, the smart traffic light units collaboratively monitor and detect vehicles coming in and out of the work zone, and then decide on when to switch the traffic direction. We design a 4-phase automated work zone traffic control algorithm based on a collaborative leader-follower strategy. Finally, we experimentally validate the auto-flagger system on our connected and autonomous vehicle proving ground.