Railway transportation is important modern infrastructure, and video surveillance systems are frequently used to ensure safety and efficiency. However, intrusion object detection from surveillance images faces significant challenges under low-light condition, which is common during nighttime or within tunnel environment. These low-illumination conditions often result in elevated noise levels, diminished contrast ratios, and loss of critical visual details, thereby impairing the performance of conventional detection algorithms. To address these issues, an enhanced YOLOv8-based algorithm model that incorporates a triple attention mechanism is proposed to improve the accuracy and efficiency of intrusion object detection in low-light environments. We first introduce a comprehensive railway-specific dataset that includes self-collected railway surveillance images and relevant classes from the EXDark dataset. The triple attention mechanism that synergistically combines channel, spatial and temporal attention for improved feature representation is then presented. The experimental results show that the proposed model demonstrates real-time processing capabilities, meeting the stringent operational response requirements of railway applications.