An algorithm for open railway obstacle intrusion detection

What the paper says

Object detection is a key technology in the field of open railway obstacle detection. However, due to limitations in onboard computational resources, real-time requirements, and complex environmental factors, traditional object detection methods are difficult to implement for efficient and accurate obstacle detection in railway systems. To address these challenges, this paper proposes an open railway obstacle detection method based on YOLOv11. First, the RepViT module is introduced to replace the backbone network of the original model, reducing the model’s structural complexity and size. Second, the spatial and channel synergistic attention mechanism is integrated into the cross-stage partial spatial attention module to enhance the network’s ability to extract spatial and channel attention features. In addition, the iterative attentional feature fusion mechanism is incorporated to redesign the original C3K2 module, enabling more effective fusion of features from different scales and semantics. Experimental results show that, compared with the original model, the improved YOLOv11 achieves a 25.2$\%$ reduction in the number of parameters and an 18.8$\%$ decrease in computational cost while maintaining detection accuracy. Moreover, the real-time video detection speed on embedded platforms reaches 25.4 frames per second (FPS). Finally, the applicability and real-time performance of the proposed algorithm is validated on the Jetson Xavier NX, providing accurate decision support for open railway safety applications.