Abstract: Multispectral pedestrian detection models use complementary enhancement of visible and infrared images to improve robustness and reliability. However, effectively fusing different modal information to reduce the leakage rate remains a challenge in multispectral pedestrian detection. To address this, we proposed a multispectral pedestrian detection model based on Dempster-Shafer theory. First, a multi-scale cross-modal homogeneous feature enhancement model was designed to enhance the features of a single modality.Then, the Dirichlet distribution was used to characterize the uncertainty of pedestrian detection, thus improving the accuracy of detection. Finally, the detection results of multiple modalities were credibly integrated using the Dempster-Shafer evidence theory, which could reduce the impact of damaged modalities on the detection results. The experimental results show that the performance of the proposed model on the KAIST dataset is better than that of the existing multi-spectral pedestrian detection model, and the log-average miss rate (MR⁻²) on the 3 major subsets of all-day, day and night is 1.84%, 1.23% and 2.16% higher than that of the suboptimal algorithm, respectively.