Enhanced line-structured light 3D object detection based on a multimodal noise identification with feature fusion

Three-dimensional detection based on line-structured light captures spatial information by projecting structured light and analyzing its deformation; however, point cloud data often suffer from noise caused by reflections and complex geometries, which impedes accurate target recognition. This paper proposes a multimodal noise identification network to effectively remove small-scale noise and improves the FCAF3D framework by incorporating a CBAM-based multiscale feature fusion module and optimizing the loss function with CIOU. Experimental results demonstrate that the proposed method significantly enhances point cloud denoising, improves detection precision, increases inference speed, and yields smoother 3D reconstructed surfaces.