Estimation of left behind subway passengers through archived data and video image processing

Area of Science:

• Transportation Engineering
• Computer Vision
• Operations Research

Background:

• Public transportation systems worldwide face significant crowding challenges.
• Extreme crowding results in passengers being left behind, unable to board trains or buses.
• Existing methods for passenger counting (e.g., farecard data) are insufficient for platform-level analysis.

Purpose of the Study:

• To develop and apply a novel methodology for estimating the number of passengers left behind on subway platforms.
• To leverage emerging object detection technology combined with existing data sources for accurate passenger counting.
• To address the limitations of current passenger counting methods in subway systems.

Main Methods:

• Utilized object detection software on surveillance video feeds to count passengers on platforms.
• Combined automatically counted passengers with train operations data.
• Developed and calibrated logistic regression models using manual counts for validation.

Main Results:

• The methodology was successfully applied to the Boston subway system, focusing on North Station during peak hours.
• Analysis of inferred origin-destination data identified stations with a high likelihood of passengers being left behind.
• The fused data approach accurately estimated the number of left-behind passengers, with results presented for a typical weekday.

Conclusions:

• Fusing passenger counts from video analysis with train operations data provides a reliable method for estimating left-behind passengers.
• The developed models demonstrate the potential for improving the management of crowded public transportation.
• This approach offers a significant advancement in understanding and mitigating the impact of extreme crowding on subway platforms.