Detecting, Tracking and Counting People Getting On/Off a Metropolitan Train Using a Standard Video Camera

Area of Science:

• Transportation Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Station delays significantly reduce public transport commercial speed.
• Manual data collection for passenger flow is time-consuming and prone to errors.
• Existing video surveillance systems offer potential for automated analysis.

Purpose of the Study:

• To develop an automated method for collecting and analyzing passenger data at public transport stations.
• To improve the accuracy and efficiency of data used for traffic modeling and experimentation.
• To leverage intelligent sensing and deep learning for enhanced public transport operations.

Main Methods:

• Utilized a new public video dataset from real-scale laboratory recordings.
• Annotated video data with head locations to create a ground-truth dataset.
• Trained and evaluated deep learning detection and tracking algorithms for passenger counting.

Main Results:

• Achieved a mean accuracy of 92% in counting passengers boarding and alighting.
• Demonstrated a standard deviation of less than 0.15% on unseen video sequences.
• Successfully exploited intelligent sensing for automated passenger flow analysis.

Conclusions:

• Automated passenger counting using deep learning significantly enhances data collection for transport analysis.
• The developed method offers a scalable solution to reduce station delays and improve public transport efficiency.
• Intelligent sensing provides a viable approach to overcome limitations of traditional data collection methods.