G-VOM++: precision off-road obstacle detection via dual-LiDAR fusion and plane calibration

The point cloud-based G-VOM obstacle detection method faces significant challenges in off-road environments, primarily due to point cloud tilting caused by vehicle shaking. This misalignment between height maps across consecutive frames adversely affects the accuracy of obstacle detection. To address this issue, G-VOM++ adopts a horizontal plane calibration technique to optimize the height map, eliminating the effects of point cloud tilt induced by vehicle motion and enabling accurate representation of the driving surface elevation. After calibration, G-VOM++ achieves a 6.55 percentage point improvement in recall rate, effectively reducing both missed detections and false positives. Moreover, in off-road environments, the presence of trees and shrubs adds further complexity to detection. To overcome this challenge, G-VOM++ incorporates the Off-Fusion fusion strategy, which enhances obstacle detection performance. By fusing data from dual LiDAR sensors, it increases both data diversity and observational perspectives, further improving detection accuracy. Experimental results show that, compared with the original method, the GVOM++ approach achieves improvements of 11.96 percentage points in precision and 10.89 percentage points in recall rate.