Point cloud change detection with global-local features fusion

Due to the rapid development of autonomous driving in recent years, point cloud change detection has attracted considerable attention due to its crucial role in enabling accurate environmental understanding. However, it is difficult to fully extract point features for 3D point cloud change detection. Prior works often prioritize local point features while overlooking the importance of global contextual information, which is essential for understanding the overall structure of the scene. To address this issue, we propose a novel Dense Fusion Network (DFN) that effectively integrates point-wise local features with global context information, enhancing the detection of changes in the point cloud. In this network, we leverage point convolution to extract local features, which are then fused with the global context using the attention mechanism. The experiments show that DFN achieves a 30% improvement in OA compared to existing methods on two datasets.