This research paper delves into the utilization of a Kinect-based algorithm for human action recognition to assess the quality of cardiopulmonary resuscitation (CPR) procedures. By leveraging Kinect's skeleton tracking capabilities, the algorithm extracts critical compression depth (CCD) and compression frequency (CCF) data, enhancing the efficiency of CPR quality evaluation by 60% compared to existing methods. This innovative approach not only ensures real-time performance and accuracy but also provides a feasible and effective strategy for CPR training in diverse settings, exhibiting promising implications for the intersection of computer vision and medical training.

Introduction

Video-based human action recognition in computer vision has gained significant popularity due to its utilization in various applications like video surveillance and human-computer interaction. The introduction highlights the significance of Kinect sensor camera in analyzing human actions through skeleton tracking capabilities, enabling researchers to study human motion trajectories effectively.

Key points included:

• The Kinect sensor camera by Microsoft revolutionized human action analysis by providing powerful skeleton tracking features.
• In China, a significant number of deaths occur due to cardiac arrest, emphasizing the importance of effective CPR, especially in non-hospital settings.
• Effective CPR within 3-5 minutes can significantly increase the survival rate of patients.
• The lack of CPR training and standardized procedures poses challenges in timely and effective resuscitation.
• Traditional CPR training methods using expensive dummies lack accessibility to the general public and feedback on trainee performance.
• The proposed algorithm in this paper leverages Kinect's capabilities to provide real-time analysis and correction of CPR procedures, enhancing efficiency, accuracy, and stability.
• The algorithm aims to improve the accessibility and effectiveness of CPR training and evaluation.

Related Work

The field of video-based human action recognition has seen significant progress with high recognition accuracy, particularly focused on color video understanding. Key points include:

• Researchers have worked extensively on video-based human action recognition, leveraging color video understanding for accurate recognition.
• Convolutional neural networks have played a crucial role in understanding image information, with continual enhancements from traditional models.
• Karen et al. introduced two-stream network architectures to extract appearance and motion features using color images and optical flow data.
• Swathikiran et al. developed a system that necessitates marker placement on trainers' hands, which comes with limitations.
• Li et al. highlighted the importance of professional CPR training, noting the challenges untrained bystanders face in providing effective CPR during drowning events.
• Xu et al. integrated Kinect equipment with fitness exercises, incorporating wearable technology for real-time human movement data collection, guiding training actions, correcting posture, and enabling effective human-computer interaction.

Proposed CPR Training Model