BFMF-Net: Bimodal Feature Multiscale Fusion Network for Multilayer Steel Bar Intersection Detection

At present, steel bar binding robots solve the problems of large amounts of work and worker safety in steel bar binding projects, and the detection of steel bar binding intersections has become a research point. To solve the problems of wrong detection and unsatisfactory detection accuracy of intersections in the lower non-bound layer due to the multi-layer structure of rebar and the complex environment at the site, this paper proposes a bimodal feature multiscale fusion network, BFMF-Net. To better fuse the RGB and depth map features, a bimodal feature fusion module, BFFM, and a multiscale fusion module, MFM, are proposed. To validate the effectiveness of the network, the validation results on the val dataset and the test dataset of the constructed bimodal multilayer steel binding dataset Bi-MSB indicate that, relative to the results of the base algorithm for detecting the intersections on the RGB alone, the detection accuracies of the proposed network are improved by 0.005 and 0.027, respectively, and the robustness to complex environments is also significantly improved.