Human poses recognition based on Spiking Pulse Graph Neural Networks.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Temporal dilated convolutional models for human pose recognition are computationally intensive.
• High complexity in input images leads to low accuracy and significant energy consumption in existing models.
• There is a need for more efficient and accurate human pose recognition methods.

Purpose of the Study:

• To design a novel Spiking Pulse Graph Neural Networks (SPGNN) model for human pose recognition.
• To enhance feature extraction accuracy and reduce model energy consumption.
• To improve overall performance compared to existing state-of-the-art methods.

Main Methods:

• Modified the receptive field processing module by adjusting convolution expansion coefficients and activation functions.
• Implemented an SPGNN to control feature information transmission rates.
• Utilized a reward mechanism based on human pose learning rates to optimize the model.

Main Results:

• The proposed SPGNN model demonstrated improved accuracy in human pose recognition.
• The model achieved a significant reduction in energy consumption.
• Performance gains were validated against temporal dilated convolutional models and other leading methods on the same test dataset.

Conclusions:

• The SPGNN model offers a more efficient and accurate solution for human pose recognition tasks.
• Adjustments to the receptive field and controlled information transmission are key to enhanced performance.
• This approach effectively addresses the limitations of high computational cost and energy demands in complex human pose estimation.