Barefoot Footprint Detection Algorithm Based on YOLOv8-StarNet.

Area of Science:

• Computer Vision
• Biometric Identification
• Machine Learning

Background:

• Conventional image recognition struggles with barefoot footprint data due to concentrated features and rich textures.
• Existing methods lack efficiency in processing complex footprint patterns for security and forensic applications.

Purpose of the Study:

• To develop an optimized footprint recognition model for enhanced biometric identification.
• To improve accuracy and efficiency in analyzing barefoot footprint images for security, medical, and criminal investigation.

Main Methods:

• Integration of an enhanced StarNet architecture into the YOLOv8 backbone for efficient feature extraction.
• Utilizing element-wise multiplication for mapping inputs to a high-dimensional nonlinear feature space.
• Employing an Encoder layer with multi-scale feature fusion and attention mechanisms for semantic information extraction.
• Implementing a feature modulation block to synergistically combine global and local information, reducing redundancy and computational complexity.

Main Results:

• Achieved a recognition accuracy of 99.5% on a proprietary barefoot footprint dataset.
• Reduced model parameters by 0.73 million, enhancing processing speed.
• Decreased GFLOPS by 1.5, lowering hardware performance requirements.
• Demonstrated significant advantages in parameter efficiency, recognition accuracy, and computational complexity.

Conclusions:

• The proposed enhanced StarNet-YOLOv8 model offers a lightweight and highly accurate solution for barefoot footprint recognition.
• The model's efficiency and accuracy make it suitable for real-world applications in biometrics and forensics.
• Future work will focus on extending capabilities to shoeprint analysis in complex forensic scenarios.