MFA-Net: Motion Feature Augmented Network for Dynamic Hand Gesture Recognition from Skeletal Data.

Area of Science:

• Computer Vision
• Machine Learning
• Human-Computer Interaction

Background:

• Dynamic hand gesture recognition is crucial for intuitive human-computer interaction.
• Existing methods often struggle to capture complex articulated movements.
• Skeletal data offers a compact representation for gesture analysis.

Purpose of the Study:

• To propose a novel network, MFA-Net, for dynamic hand gesture recognition using skeletal data.
• To enhance feature representation by incorporating both finger and global motion patterns.
• To improve the accuracy and robustness of dynamic hand gesture recognition systems.

Main Methods:

• Developed a Motion Feature Augmented Network (MFA-Net) integrating skeletal data with motion features.
• Extracted finger motion features using a variational autoencoder to capture articulated movements.
• Utilized global motion features for overall hand movement representation.
• Employed a recurrent neural network (RNN) with three branches to process and augment features.

Main Results:

• MFA-Net demonstrated comparable performance on the DHG-14/28 dataset.
• The proposed method achieved superior performance on the SHREC'17 dataset compared to state-of-the-art techniques.
• Augmenting features with motion information significantly improved classification accuracy.

Conclusions:

• MFA-Net effectively recognizes dynamic hand gestures from skeletal data by leveraging novel motion features.
• The integration of finger and global motion features enhances deep network representations for improved gesture recognition.
• This approach offers a promising direction for advancing human-computer interaction through more accurate gesture understanding.