Active Localization of Noncooperative UAV Based on Gimbaled LiDAR--Camera Fusion

Abstract

Robust 3-D localization and tracking of a noncooperative uncrewed aerial vehicle (UAV) is fundamentally constrained by the tradeoff between wide field-of-view (FoV) coverage and high-resolution, long-range 3-D sensing. Although solid-state light detection and ranging (LiDAR) sensors provide high-resolution and long-range sensing, the narrow FoV often leads to target loss during large-amplitude UAV maneuvers. To mitigate this limitation, this article develops an active sensing framework that integrates a solid-state LiDAR and a monocular camera on a 2-DOF gimbal for continuous single-target UAV localization and tracking, but the dynamic observation introduces two major challenges: rapid gimbal motion causes severe intraframe point-cloud distortion, and long-range sensing yields sparse LiDAR returns together with fluctuating visual detections, making data association unstable. To address these challenges, we propose pointwise LiDAR motion compensation based on tightly coupled inertial measurement unit (IMU) and encoder fusion, 3-D candidate generation with interacting multiple model (IMM)-based motion-frustum dual priors, and a probabilistic data association filter with soft visual confidence and an empty-association hypothesis. Experiments including distortion compensation tests and 16 outdoor UAV flights show effective compensation of gimbal-induced distortion, mean per-flight 3-D RMSE/P95 values of 0.528/0.849 m on confirmed-valid online outputs, and continuous tracking over evaluated ranges of 9.94–51.55 m with the current hardware. The results indicate that active gimbal steering expands the effective observation sector beyond the instantaneous FoV of the LiDAR.