The Infrared- Visible Complementary Recognition Network Based on Context Information

In recent years, object recognition has received more and more attention, but how to effectively overcome the visual ambiguity problem in the bad environment such as low illumination is still a very challenging problem. On the other hand, we consider the context information of the object to provide extra discriminative information. Therefore, multi-modal recognition based on the strong correlation between different parts of the target has the potential to solve the above problems. In this work, we propose a novel infrared-visible complementary recognition network (IV-CRN) to solve these problems. The IV-CRN includes a fusion branch and a global branch. The fusion branch uses a brightness-based weighted fusion algorithm to fuse the complementary information in the bimodal state and the convolution gated recurrent unit (ConvGRU) to extract target context information. The global branch extracts global information and complements the fragmented information of the fusion branch. We conducted comparative experiments and ablation experiments on two challenging datasets. The experimental results show that our method has good performance compared with the existing multi-mode object recognition methods. Compared with the method based on single modal image, this method also significantly improves the object accuracy.