passenger recapture across multiple cameras
- Project
- 21032 TAPCOP
- Type
- New library
- Description
The project delivered an AI pipeline that reliably tracks passengers as they move between CCTV camera views onboard trains. This is challenging due to low camera angles, the linear layout of trains, dynamic lighting, occlusion by seats or other passengers, and changes in passenger appearance (e.g., removing jackets). The solution addresses these complexities, enabling long-term passenger recapture across distributed cameras. This result is a key advancement for applications like passenger counting, flow monitoring, and left luggage detection, enhancing train safety and operational efficiency.
- Contact
- Steven Lauwereins; Televic Rail NV
- s.lauwereins@televic.com
- Research area(s)
- AI, CCTV
- Technical features
The project delivered an AI-based pipeline for robust passenger tracking across multiple CCTV cameras in trains. Key technical features include passenger recapturing based on the known camera layout, enabling seamless tracking across distributed cameras. The system handles gradual changes in passenger appearance, such as removing jackets or adding accessories, and supports long-term recapturing over several hours. Advanced dynamic lighting compensation ensures reliable operation in challenging conditions, such as transitions through tunnels or direct sunlight. This solution significantly enhances capabilities for applications like passenger flow monitoring and left luggage detection.
- Integration constraints
Currently the solution requires high-end GPU's for real-time processing. The solution is an AI-based pipeline
- Targeted customer(s)
Train Operating Companies, Train Builders
- Conditions for reuse
Televic Rail is sole owner of the IP
- Confidentiality
- Public
- Publication date
- 01-07-2024
- Involved partners
- SIRRIS (BEL)
- Televic Rail N.V. (BEL)