A Combination Model of Shifting Joint Angle Changes With 3D-Deep Convolutional Neural Network to Recognize Human...

Area of Science:

• Computer Science
• Biomedical Engineering
• Robotics

Background:

• Human activity recognition (HAR) is crucial for applications like medical rehabilitation.
• Current HAR methods struggle to differentiate activities with similar joint distance changes but different motion directions.
• There is a need for advanced HAR techniques to accurately detect and respond to diverse human movements.

Purpose of the Study:

• To develop an innovative approach for human activity recognition that accurately identifies movement direction.
• To overcome limitations of existing methods in distinguishing activities with similar joint distance variations.
• To enhance the efficiency and accuracy of HAR systems for applications in medical rehabilitation and beyond.

Main Methods:

• A novel joint angle shift approach was developed to analyze movement direction.
• The joint angle shift method was integrated with a Deep Convolutional Neural Network (DCNN) model.
• The DCNN model processed 3D datasets with spatio-temporal information from RGB-D video data.

Main Results:

• The model successfully classified nine activities, including sitting and standing, on the Florence 3D Actions dataset with 96.72% accuracy.
• Testing on the UTKinect Action3D dataset yielded an accuracy of 97.44%, demonstrating robust performance.
• The joint angle shift method effectively distinguished activities with similar joint distance changes but different motion directions.

Conclusions:

• The proposed joint angle shift method combined with DCNN achieves state-of-the-art performance in human activity recognition.
• This approach offers a significant advancement in accurately recognizing human movements, particularly in complex scenarios.
• The findings have strong implications for improving HAR systems in medical rehabilitation and other related fields.