Multimodal 3D target detection based on improved Mvxnet

To enhance the feature extraction and feature fusion ability of the model and further improve the detection effect of the model, this paper designs a loop feature fusion module based on the Mvxnet network, which aims to improve the feature fusion ability of the model after feature extraction from RGB images and to improve the feature extraction ability of the model for objects of different sizes. A fusion mechanism of RGB image features and 3D point cloud weighted features is designed, which can adaptively fuse useful target features and suppress invalid features from RGB graphics and 3D point clouds, further improving the model’s feature processing ability. Finally, after the fusion feature information is obtained, further to improve the final feature processing ability of the model, this paper designs a spatial feature fusion module to obtain spatial attention information from the feature map in the upper layer. It is fused to another feature graph to get the attention of the feature graph. To verify the performance of the model, experiments are carried out on KITTI data sets. The experimental results show that the improved strategy proposed in this paper improves the detection effect of the model while ensuring that the detection speed is unchanged. It shows that the improved algorithm proposed in this paper has important research value and practical significance.