Multi-level Attention Network for Accurate Crowd Counting in Challenging Rail Transit Environment

Key Points

• Accurate crowd counting is achieved using a multi-level attention network, enhancing identification of dense crowds.
• The model shows a remarkable performance with a mean absolute error of 3.5 people in crowded scenarios.
• Evaluation utilized a dataset from various rail transit environments characterized by diverse crowd behaviors and patterns.
• Improvements in accuracy may enable better safety management in public transportation systems, particularly during peak times.