Unmanned Aerial Vehicle Path Planning for Traffic Estimation and Detection of Non-Recurrent Congestion

Abstract

Verbatim from arXiv. Not paraphrased, not summarized.

Unmanned aerial vehicles (UAVs) provide a novel means of extracting road and traffic information from video data. In particular, by analyzing objects in a video frame, UAVs can detect traffic characteristics and road incidents. Leveraging the mobility and detection capabilities of UAVs, we investigate a navigation algorithm that seeks to maximize information on the road/traffic state under non-recurrent congestion. We propose an active exploration framework that (1) assimilates UAV observations with speed-density sensor data, (2) quantifies uncertainty on the road/traffic state, and (3) adaptively navigates the UAV to minimize this uncertainty. The navigation algorithm uses the A-optimal information measure (mean uncertainty), and it depends on covariance matrices generated by a dual state ensemble Kalman filter (EnKF). In the EnKF procedure, since observations are a nonlinear function of the incident state variables, we use diagnostic variables that represent model predicted measurements. We also present a state update procedure that maintains a monotonic relationship between incident parameters and measurements. We compare the traffic/incident state estimates resulting from the UAV navigation-estimation procedure against corresponding estimates that do not use targeted UAV observations. Our results indicate that UAVs aid in detection of incidents under congested conditions where speed-density data are not informative.

Authors

• Cesar N. Yahia
• Shannon E. Scott
• Stephen D. Boyles
• Christian G. Claudel

Keywords

• stat.AP
• eess.SY