Non-Line-of-Sight Perception Method for Autonomous Haul Trucks in Open-Pit Mines Based on 4D mmWave Radar and LiDAR Fusion.

In open-pit mining environments, large equipment frequently causes severe occlusion, creating critical perception blind spots for LiDAR. Meanwhile, 4D millimeter-wave (mmWave) radar provides penetration capability but is highly susceptible to multipath interference generated by metallic structures and uneven terrain. To address these challenges, this paper proposes a Blind-Spot Complementary Fusion (BSCF) framework that integrates 3D LiDAR and 4D mmWave radar through explicit geometric constraints. The proposed framework first suppresses multipath artifacts and performs calibrated spatiotemporal alignment between heterogeneous sensors. It then introduces high-confidence radar observations into LiDAR blind spots through spatial consistency verification, providing existence-level hidden-target risk cues under extreme occlusion. In addition, a Volume Recovery Rate (VRR) proxy metric is proposed to quantitatively describe envelope-level spatial evidence in occluded regions. Experiments conducted on real-world mining datasets demonstrate that the proposed method effectively suppresses severe multipath interference and improves scene-level cross-modal proximity by approximately 16.5%, while the overlap-region RMSE remains approximately 0.18 m. Under complete occlusion where LiDAR observations are unavailable, the framework provides target-existence risk cues with a VRR of up to 15.6%, providing a useful preliminary indication of hidden target existence. Ultimately, the proposed approach enhances perception robustness and safety for autonomous transportation systems in open-pit mines.