Face detection for rail transit passengers based on single shot detector and active learning.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Public Health Surveillance

Background:

• COVID-19 necessitates mask-wearing in public spaces like rail transit.
• Existing face detection struggles to differentiate masked from unmasked individuals.
• Accurate face and mask detection is crucial for transit safety and health monitoring.

Purpose of the Study:

• To develop a robust face detection algorithm for rail transit surveillance.
• To effectively detect both faces and faces wearing masks in real-time.
• To enhance the generalization and efficiency of face detection models using active learning.

Main Methods:

• Implemented a real-time face detection algorithm using Single Shot Detector (SSD).
• Optimized the SSD algorithm by refining the backbone network, feature pyramid network, spatial attention module, and loss function.
• Employed a semi-supervised active learning strategy to select informative samples for retraining, improving model generalization and reducing labeling efforts.

Main Results:

• The proposed algorithm demonstrated significant performance improvements on a rail transit dataset.
• Achieved superior accuracy in detecting both masked and unmasked faces compared to state-of-the-art methods.
• The active learning approach effectively reduced the need for extensive manual data labeling.

Conclusions:

• The developed face detection algorithm is highly effective for rail transit surveillance.
• Applications include passenger flow analysis, epidemiological studies, and enforcing mask mandates.
• The system prioritizes passenger privacy by not collecting or storing facial data.