Multi-LiDAR human joint recognition algorithm in hospital wards based on improved V2V-Posenet

Area of Science:

• Medical technology
• Computer vision
• Robotics

Background:

• Patient safety in hospitals necessitates continuous monitoring, especially for preventing falls and convulsions when staff are absent.
• Existing human joint recognition methods face challenges like occlusion and complex backgrounds, limiting their real-world application.

Purpose of the Study:

• To develop a robust human joint detection system for hospital wards using a multi-LiDAR approach.
• To improve the accuracy and efficiency of patient monitoring systems for enhanced safety.

Main Methods:

• A multi-LiDAR system was employed to capture multi-view human body point clouds.
• An improved V2V-Posenet model was utilized for precise human joint position detection.
• Techniques including point cloud splicing, voxelization, data enhancement, loss function constraints, and 3D Gaussian filtering were implemented.

Main Results:

• The improved V2V-Posenet model achieved a mean average precision of 91.6%, a significant increase from the original model's 80.1%.
• Performance surpassed the comparison algorithm A2J-Posenet (77.4% mAP).
• The model demonstrated real-time target detection capabilities, meeting practical requirements.

Conclusions:

• The proposed multi-LiDAR system and improved V2V-Posenet model offer a highly effective solution for human joint detection in hospital environments.
• This technology can enhance patient safety by enabling reliable monitoring and timely intervention.
• The system's accuracy and speed make it suitable for practical deployment in healthcare settings.