PREACT – Predictive Analytics for Crowd Transit Optimization

AI Quick Summary

Our project aims to optimize passenger flows at train stations by combining computer vision analysis from static cameras for indoor spaces with drone-based data collection for outdoor areas. In collaboration with EPFL’s VITA Lab (Prof. Alahi - academic partner), we will track passenger origin-destination flows, including connections to bike/scooter stations and other public transport systems such as metro and buses. This privacy-friendly approach will help identify conflicts in walking trajectories, high-density areas on platforms, and walking times during interconnections. Additionally, we will leverage predictive analytics to forecast crowd density based on historical data, weather conditions, and scheduled events, enabling SBB to proactively manage congestion, improve safety and passenger transit. The project is highly scalable, allowing deployment at any station in Switzerland or abroad. Long DescriptionEfficient passenger movement is crucial for enhancing safety, comfort, and operational efficiency at train stations. In collaboration with EPFL’s VITA Lab (Prof. Alexandre Alahi), this project will integrate computer vision analysis from static cameras (focused on indoor station areas) with drone-based data collection (capturing outdoor pedestrian flows). The fusion of these data sources will enable a detailed, privacy-friendly analysis of crowd movement, solely measuring traffic patterns and density without identifying personal information.By tracking where passengers come from and where they go after leaving the train — including their connections to bike/scooter stations, metro, and buses—the project will open up new opportunities for strategic collaborations and incentives with mobility service providers. Bike and e-scooter sharing companies could offer targeted discounts or priority access at high-traffic stations, encouraging seamless multimodal travel. Additionally, real estate businesses and retail operators near stations could benefit from hotspot analysis of pedestrian flow, allowing them to adjust their services dynamically. For instance, rather than waiting on the platform, passengers might be encouraged to wait at a nearby café or take advantage of take-away restaurant services, improving both station efficiency and the passenger experience.From a safety perspective, traditional accident data is often unavailable or incomplete, making it difficult to assess risk accurately. To address this, the project will implement Surrogate Safety Measures (SSMs)—a well-established methodology in transportation research that identifies near-misses, pedestrian conflicts, and high-density areas as early indicators of safety risks. By analyzing movement patterns, sudden stops, evasive actions, trajectory conflicts and proximity to platform edges, we can proactively detect and mitigate safety risks before they result in actual incidents.Key ObjectivesOptimize passenger movement by identifying conflicts in walking trajectories, reducing congestion, and ensuring smooth transitions between transport modes.Improve safety by analyzing high-density areas on platforms, where crowding poses risks.Enhance the travel experience by minimizing unnecessary waiting times and improving station design based on observed movement patterns.Provide SBB with SSMs for pedestrians, filling the gap left by the lack of accident data and enabling proactive risk management.Leverage predictive analytics to forecast crowd density at platforms and stations by analyzing historical data, weather conditions, and scheduled events, allowing for proactive congestion management.Prioritize high-impact scenarios, such as rush hour, when congestion and operational challenges peak.Roles & ResponsibilitiesMobiLysis (EPFL spin-off): Responsible for drone-based data collection and integration with static camera data. The approach prioritizes offline and simulation-based analysis over real-time implementation, ensuring higher-quality insights for long-term decision-making rather than short-term reactive measures.VITA (EPFL): Conducts computer vision analysis on static camera footage, fusing it with drone data to track passenger flows and analyze crowd movement challenges such as conflicts, density, and walking times.A VITA master's student, co-supervised by a MobiLysis team member, will execute the research and analysis tasks.Key Performance Indicators (KPIs)Passenger waiting times before train departure, based on their means of arrival.Origin-destination flows based on train lines and intermodal transitions.Safety surrogate measures (SSMs), such as crowd density and movement conflicts on platforms.Enhanced crowd management strategies, particularly during peak hours and special events.Accuracy and effectiveness of predictive analytics in forecasting crowd density and passenger movement patterns.Impact & ScalabilityThe results of this project will provide SBB with actionable insights to improve station design, reduce congestion risks, and enhance service quality for passengers. Furthermore, these insights can help form strategic partnerships that promote seamless mobility across different transport modes. Our solution is highly scalable and ready for application, meaning it can be deployed at any station with static camera infrastructure, ensuring long-term value and adaptability across the SBB network.9-Month Project PlanM1 - Project kickoff, finalizing methodology, defining station test sitesM2 - Data collection planning, obtaining necessary permits for drones and camera accessM3 - Deployment of static cameras, drone test flights, initial data acquisitionM4-5 - Computer vision model refinement and calibration, initial pedestrian flow analysisM6-8 - Integration of static and drone data, development of safety surrogate measures (SSMs)M9 - Final data analysis, predictive modeling, and drafting of key insights for SBB

Project Idea Metadata

• Project Idea Name: PREACT – Predictive Analytics for Crowd Transit Optimization
• Date: 3/6/2025 12:42:43 PM
• Administrators:
  • Manos Barmpounakis

Project Idea Description

Long Description

Efficient passenger movement is crucial for enhancing safety, comfort, and operational efficiency at train stations. In collaboration with EPFL’s VITA Lab (Prof. Alexandre Alahi), this project will integrate computer vision analysis from static cameras (focused on indoor station areas) with drone-based data collection (capturing outdoor pedestrian flows). The fusion of these data sources will enable a detailed, privacy-friendly analysis of crowd movement, solely measuring traffic patterns and density without identifying personal information.

By tracking where passengers come from and where they go after leaving the train — including their connections to bike/scooter stations, metro, and buses—the project will open up new opportunities for strategic collaborations and incentives with mobility service providers. Bike and e-scooter sharing companies could offer targeted discounts or priority access at high-traffic stations, encouraging seamless multimodal travel. Additionally, real estate businesses and retail operators near stations could benefit from hotspot analysis of pedestrian flow, allowing them to adjust their services dynamically. For instance, rather than waiting on the platform, passengers might be encouraged to wait at a nearby café or take advantage of take-away restaurant services, improving both station efficiency and the passenger experience.

From a safety perspective, traditional accident data is often unavailable or incomplete, making it difficult to assess risk accurately. To address this, the project will implement Surrogate Safety Measures (SSMs)—a well-established methodology in transportation research that identifies near-misses, pedestrian conflicts, and high-density areas as early indicators of safety risks. By analyzing movement patterns, sudden stops, evasive actions, trajectory conflicts and proximity to platform edges, we can proactively detect and mitigate safety risks before they result in actual incidents.

Key Objectives

• Optimize passenger movement by identifying conflicts in walking trajectories, reducing congestion, and ensuring smooth transitions between transport modes.
• Improve safety by analyzing high-density areas on platforms, where crowding poses risks.
• Enhance the travel experience by minimizing unnecessary waiting times and improving station design based on observed movement patterns.
• Provide SBB with SSMs for pedestrians, filling the gap left by the lack of accident data and enabling proactive risk management.
• Leverage predictive analytics to forecast crowd density at platforms and stations by analyzing historical data, weather conditions, and scheduled events, allowing for proactive congestion management.
• Prioritize high-impact scenarios, such as rush hour, when congestion and operational challenges peak.

Roles & Responsibilities

• MobiLysis (EPFL spin-off): Responsible for drone-based data collection and integration with static camera data. The approach prioritizes offline and simulation-based analysis over real-time implementation, ensuring higher-quality insights for long-term decision-making rather than short-term reactive measures.
• VITA (EPFL): Conducts computer vision analysis on static camera footage, fusing it with drone data to track passenger flows and analyze crowd movement challenges such as conflicts, density, and walking times.
• A VITA master's student, co-supervised by a MobiLysis team member, will execute the research and analysis tasks.

Key Performance Indicators (KPIs)

1. Passenger waiting times before train departure, based on their means of arrival.
2. Origin-destination flows based on train lines and intermodal transitions.
3. Safety surrogate measures (SSMs), such as crowd density and movement conflicts on platforms.
4. Enhanced crowd management strategies, particularly during peak hours and special events.
5. Accuracy and effectiveness of predictive analytics in forecasting crowd density and passenger movement patterns.

Impact & Scalability

The results of this project will provide SBB with actionable insights to improve station design, reduce congestion risks, and enhance service quality for passengers. Furthermore, these insights can help form strategic partnerships that promote seamless mobility across different transport modes. Our solution is highly scalable and ready for application, meaning it can be deployed at any station with static camera infrastructure, ensuring long-term value and adaptability across the SBB network.

9-Month Project Plan

• M1 - Project kickoff, finalizing methodology, defining station test sites
• M2 - Data collection planning, obtaining necessary permits for drones and camera access
• M3 - Deployment of static cameras, drone test flights, initial data acquisition
• M4-5 - Computer vision model refinement and calibration, initial pedestrian flow analysis
• M6-8 - Integration of static and drone data, development of safety surrogate measures (SSMs)
• M9 - Final data analysis, predictive modeling, and drafting of key insights for SBB

Our project aims to optimize passenger flows at train stations by combining computer vision analysis from static cameras for indoor spaces with drone-based data collection for outdoor areas. In collaboration with EPFL’s VITA Lab (Prof. Alahi - academic partner), we will track passenger origin-destination flows, including connections to bike/scooter stations and other public transport systems such as metro and buses. This privacy-friendly approach will help identify conflicts in walking trajectories, high-density areas on platforms, and walking times during interconnections. Additionally, we will leverage predictive analytics to forecast crowd density based on historical data, weather conditions, and scheduled events, enabling SBB to proactively manage congestion, improve safety and passenger transit. The project is highly scalable, allowing deployment at any station in Switzerland or abroad.