Search papers, labs, and topics across Lattice.
This paper introduces the Viking Hill Dataset, a novel multi-sensor dataset that integrates high-resolution FMCW imaging radar, lidar, and RGB camera data for enhanced detection and segmentation of forest scenes. The dataset addresses the limitations of existing forestry datasets by providing co-registered 4D imaging radar data, which shows competitive performance in semantic segmentation tasks, particularly for dominant classes like ground and canopy. Key findings reveal that while radar performs well overall, it struggles with fine geometrical structures such as tree trunks, highlighting the strengths and weaknesses of different modalities in forest perception tasks.
Radar achieves competitive segmentation performance in forest environments, challenging the dominance of lidar for robust perception under occlusion.
Autonomous robots operating under forest canopies need robust perception of trees and surrounding vegetation across varying seasonal conditions. Existing forestry datasets provide lidar or camera data with per-tree annotations, but none include co-registered 4D imaging radar -- a modality of growing interest for its resilience to visual degradation, surface contamination, and vegetation occlusion. We introduce a multi-sensor forest dataset collected by a mobile robot equipped with a high-resolution FMCW imaging radar, lidar, RGB camera, IMU, and RTK-GNSS. The site was recorded in two sessions under contrasting vegetation states, and 3D cuboid annotations -- including per-tree diameter estimates -- provide shared semantic labels across all three perception modalities. Furthermore, we provide baseline results for semantic segmentation of the radar and lidar point clouds using MinkowskiUNet. Radar achieves IoU scores competitive with lidar for dominant classes (ground 91%, canopy 86%) while lagging on geometrically fine structures such as tree trunks (56% vs. 74%). A cross-modality analysis further compares lidar and radar trunk segmentation against an RGB detection model, and a diameter-stratified evaluation reveals how trunk segmentation quality varies with tree size. Beyond segmentation, the co-registered multi-modal data and RTK-GNSS-aided reference positioning support research in mapping, localization, and sensor fusion under canopy. The dataset and annotation tools are publicly available.
