Conceitos essenciais
CRPlace proposes a background-attentive camera-radar fusion method for accurate place recognition by focusing on stationary background features.
Resumo
I. Introduction
Place recognition is crucial for autonomous systems.
Cameras and LiDAR are commonly used sensors.
Radar remains unaffected by harsh weather conditions.
II. Methodology
CRPlace integrates camera and radar data for place recognition.
BAMG module generates a background attention mask.
BSF module facilitates spatial fusion between camera and radar features.
III. Experiments
Evaluation on the nuScenes dataset shows CRPlace outperforms other methods.
Ablation studies demonstrate the effectiveness of each module in CRPlace.
IV. Comparative Study
Comparison with state-of-the-art methods in various environmental conditions.
V. Conclusion
CRPlace improves place recognition performance by fusing camera and radar data effectively.
Estatísticas
結果は、我々のアルゴリズムが総合的なメトリックで他のベースライン手法を上回っていることを示しています(recall@1が91.2%に達する)。