Idea
Real-time visual safety platform for trams detecting rails and hazards to prevent track intrusions and accidents.
Research Paper
Core Innovation
This paper presents RailSafeNet, a novel integration of semantic segmentation and object detection tailored for tram environments. It uniquely combines SegFormer B3 and YOLOv8 models with a rule-based Distance Assessor to evaluate track intrusion risks in real time. This approach improves scene understanding accuracy compared to prior single-task models, enabling proactive safety warnings.
Market Size (TAM)
$2–10B TAM, $1–2B SAM; assumption: global urban rail safety and public transport safety system markets expanding with smart city initiatives.
Potential Customers & Pain Points
- Tram Operators Needing Enhanced Safety Systems
- Public Transport Authorities Seeking Accident Reduction
- Urban Planners Improving Tram Infrastructure Safety
Business Model
Licensing software platform to tram operators and transport authorities with optional integration and support services.
Competitive Landscape
- Siemens Mobility
- Alstom
- Hitachi Rail
Implementation Challenges
- Integration with Existing Tram Systems
- Real-time Processing Constraints
- Regulatory Approvals for Safety Systems
Validation Strategy
- Pilot deployment with a major tram operator
- Collect real-world incident and warning data
- Iterate model based on operational feedback
Research Paper Overview
RailSafeNet: Visual Scene Understanding for Tram Safety
Summary
RailSafeNet is a real-time framework that uses monocular video to enhance tram safety by detecting rails, localizing nearby objects, and assessing their risk of track intrusion through semantic segmentation, object detection, and a rule-based Distance Assessor. It achieves 65% IoU with SegFormer B3 and 75.6% mAP with YOLOv8 on the RailSem19 dataset, enabling accurate scene understanding to warn tram drivers before dangerous situations escalate.