Enhancing Railway Safety Through Fast and Economical Object Detection: A Grounding DINO Approach

Abstract

Nowadays, railway safety is becoming increasingly important due to increasing railway traffic and the potential hazards around it. In this context, this study presents a fast and economical method developed to detect people around the railway and increase safety. A custom dataset obtained using the Anafi4K drone was augmented using various image enhancement techniques (flip horizontal, crop 10%, and noise). This expanded data set better reflects real-world conditions, making the model more reliable and capable of generalization. The study adopts the Grounding DINO method and the Zero-Shot Object Detection method. This approach aims to detect people around the railway as foreign matter. Grounding DINO increases object detection accuracy by improving the learning process of the model. Zero-Shot Object Detection stands out with its ability to recognize classes that have not been previously trained, which strengthens the ability to have prior knowledge and rapid response in terms of railway safety. Among the strengths of the study, the fact that the developed method is both fast and economical stands out. This feature provides a practical solution for railway operators and security units, allowing them to use resources more effectively. In addition, it increases safety standards by detecting people around the railway and contributing to the early identification of potential dangers. This study evaluates the knowledge gained by examining similar studies on railway safety and object detection and offers a new approach by focusing on the deficiencies in the field. The results of the study may inspire further research in the field of railway safety.