Description:
<jats:p>Multispectral information fusion technology is a practical approach to enhance pedestrian detection performance in low light conditions. However, current methods often overlook the impact of illumination on modal weights and the significance of inter-modal differential information. Therefore, this paper proposes a novel illumination-aware cross-modality differential fusion (IACMDF) model. The weights of the different modalities in the fusion stage are adaptively adjusted according to the illumination intensity of the current scene. On the other hand, the advantages of the respective modalities are fully enhanced by amplifying the differential information and suppressing the commonality of the twin modalities. In addition, to reduce the loss problem caused by the importance occupied by different channels of the feature map in the convolutional pooling process, this work adds the squeeze-and-excitation attention mechanism after the fusion process. Experiments on the public multispectral dataset KAIST have shown that the average miss rate of our method is substantially reduced compared to the baseline model.</jats:p>