A Real-Time Subway Driver Action Sensoring and Detection Based on Lightweight ShuffleNetV2 Network

Area of Science:

• Computer Vision
• Artificial Intelligence
• Transportation Safety

Background:

• Subway train safety relies heavily on precise driver operations and adherence to hand actions.
• Current methods lack automated systems for verifying driver actions from surveillance data.
• Accurate driver action detection is crucial for ensuring operational integrity and safety.

Purpose of the Study:

• To propose a lightweight, two-stage model for automated subway driver action sensing and detection.
• To develop a system capable of real-time monitoring and verification of driver actions from surveillance cameras.
• To enhance the safety and efficiency of subway operations through advanced computer vision techniques.

Main Methods:

• A two-stage model comprising a driver detection network (MobileNetV2-SSDLite) and an action recognition network (improved ShuffleNetV2).
• The action recognition network integrates Spatial Enhanced Module (SEM), Improved Shuffle Units (ISUs), and Shuffle Attention Modules (SAMs).
• Development of a practical surveillance system with video-reading, main operation, and result-displaying modules.

Main Results:

• The proposed model demonstrated superior performance compared to existing methods like 3D MobileNetV1, 3D MobileNetV3, SlowFast, SlowOnly, and SE-STAD.
• The developed system successfully performs real-time action sensing and detection directly from surveillance camera feeds.
• Runtime analysis confirmed the system meets real-time detection requirements for subway operations.

Conclusions:

• The lightweight two-stage model offers an effective solution for automated subway driver action detection.
• The developed system provides a reliable tool for enhancing train safety by monitoring driver actions in real-time.
• This technology has the potential to significantly improve the oversight and safety protocols in subway transportation systems.