Heterogeneous LiDAR Early Fusion and Learned Re-Ranking Strategy for Robust Long-Term Place Recognition in Unstructured Environments

Researchers have developed MinkUNeXt-VINE++, a novel approach that improves place recognition in unstructured environments like vineyards by combining data from multiple LiDAR sensors.

MinkUNeXt-VINE++ fuses data from Livox Mid-360 and Velodyne VLP-16 LiDAR sensors and employs a learned re-ranking strategy to enhance place recognition performance. According to the study published on arxiv.org^[1], this approach achieved a 20% improvement in the Recall@1 metric compared to single-sensor methods. The TEMPO-VINE dataset, which provides heterogeneous LiDAR data across different phenological stages in vineyard environments, was used to evaluate the approach. LiDAR sensors provide detailed 3D information and are invariant to lighting conditions, making them suitable for environments with varying illumination. Another study on arxiv.org^[2] highlighted the challenges of visual place recognition in natural environments and the importance of depth as a complementary modality. The proposed MinkUNeXt-VINE++ method leverages the strengths of different LiDAR sensors to provide a more comprehensive representation of the environment, particularly in repetitive environments like vineyards.