YOLOv8-G2F: A portable gesture recognition optimization algorithm.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Hand gesture recognition (HGR) is crucial for human-computer interaction but faces challenges with high hardware costs and large model sizes.
• Deep learning has advanced HGR, but there's a need for lightweight, high-precision networks for end-to-end applications.

Purpose of the Study:

• To develop an efficient and accurate lightweight network for real-time hand gesture recognition.
• To reduce the computational and memory footprint of HGR models without sacrificing performance.

Main Methods:

• Introduced YOLOv8-G2F, an enhanced network based on YOLOv8, incorporating lightweight modules replacing traditional convolutions.
• Employed linear transformations, group convolution, and depthwise separable convolution for efficient feature extraction.
• Utilized model pruning to further reduce model size and enhance accuracy.

Main Results:

• Achieved a 99.2% recognition accuracy on the nus-ii gesture dataset.
• The resulting model size is only 2.33 MB, demonstrating significant efficiency.
• YOLOv8-G2F showed substantial improvements over existing algorithms in comparative and ablation studies.

Conclusions:

• YOLOv8-G2F offers a highly accurate and lightweight solution for hand gesture recognition.
• The proposed network effectively addresses the demand for efficient end-to-end HGR systems.
• This work paves the way for broader applications of HGR in resource-constrained environments.