Dual-Module Vision-Based Framework for Close-Proximity Real-Time Fall Detection.

Area of Science:

• Robotics
• Biomechanics
• Gerontology

Background:

• Falls are a major risk to older adults' independence.
• Sit-to-stand (STS) transitions are common moments for falls.
• Existing fall detection methods may lack real-time accuracy or efficiency.

Purpose of the Study:

• To develop a real-time fall detection system integrated with the SkyWalker assistive robot.
• To accurately classify sit-to-stand phases and detect falls using kinematic data.
• To enhance the safety of older adults during mobility transitions.

Main Methods:

• Utilized a depth camera on the SkyWalker robot for close-range motion capture.
• Employed MediaPipe to generate 3D skeletal models and extract 14 kinematic features.
• Implemented a dual-modular Support Vector Machine (SVM) classification framework for STS phase and fall detection.

Main Results:

• Achieved high accuracy in real-time classification of STS phases.
• Demonstrated effective detection of irregular motions indicative of falls.
• The system proved computationally efficient and robust to skeleton distortions.

Conclusions:

• The proposed real-time fall detection system effectively monitors STS transitions for older adults.
• Integration with the SkyWalker robot offers a promising approach for fall prevention.
• Future work will focus on proactive fall prevention strategies to ensure safer mobility.