Abstract: The working environment lighting conditions of mining unmanned vehicles are complicated, and pedestrian detection often fails to detect and misdetect, resulting in insufficient reliability of mining unmanned vehicles. In this paper, the pedestrian detection technology of unmanned mining vehicles under complex lighting environment is studied, the purpose is to promote the unmanned process of underground vehicles, ensure the safety of underground workers and improve the production efficiency of enterprises. Aiming at the low light or low light in the images collected from underground, a low light image enhancement algorithm is designed to enhance the low light image and suppress the noise in the process of light enhancement through image space conversion. Design of YOLOv3 target detection network Aiming at the improvement method of pedestrian detection for mining unmanned vehicles, the YOLOv3 pedestrian detection network for mining vehicles is proposed. On the basis of YOLOv3 detection network, the dense connection module was added first to replace the Residual connection mode to improve the feature map utilization rate, and then the CBAM module was added to refine the feature from two aspects of channel and space to strengthen the feature representation ability for small targets. Compared with the YOLOv3 target detection network, the average accuracy of the YOLOv3 pedestrian detection network for mining unmanned vehicles on low-visible images was improved by 10.26%.