Table 1.

Previous studies in vision-based grade-crossing monitoring

ReferenceMethodologyLimitations
Amin et al. (2024) Combine various YOLO models with UNet to detect traffic participants and the grade crossing area for enhanced safety1) The use of a data set with repeating images for both training and testing raises concerns about potential overfitting of the model; 2) the combination of YOLO and UNet is computationally expensive
Gong et al. (2022) Propose an enhanced few-shot learning solution for intrusion detection to mitigate data imbalance1) Although suggested solutions achieved high accuracy, they are tested using limited data, which may affect their usefulness in the long run; 2) model iterations are based on manual work, which makes it time-consuming and labor-heavy; 3) the suggested solutions are only evaluated on proprietary data sets or in a single location, raising concerns about their extensibility
Klammsteiner et al. (2023) Propose a railway track monitoring system based on Yolov5 for detection and BiSeNetv2 for segmentation to detect obstacles on various track areas
Rahman et al. (2022) Explore MobileNetv2 for obstacle detection at rail crossings through data augmentation and transfer learning
Guo et al. (2022) 1) Propose a transformer-based model for traffic detection in the occluded scenes; 2) provide performance under bad video quality such as extreme weather conditionThe proposed solutions are evaluated solely on proprietary data sets with 2,358 images which are all from a single location, raising concerns about its generalizability
Tang et al. (2023) 1) Adapt segmentation enhanced with SuBsense for background generation to detect grade-crossing trespassers; 2) training the model with weak supervised learning to mitigate the lack of annotated dataThe solution combines background generation with semantic segmentation, leading to high latency
Source(s): Created by authors

or Create an Account

Close subscription notice
Close access options