Emerging trends in gait recognition based on deep learning: a survey.

Area of Science:

• Biometrics
• Computer Vision
• Artificial Intelligence

Background:

• Gait recognition is a non-invasive biometric method valued for long-distance capture and impersonation resistance.
• Deep learning has revolutionized gait recognition by enabling complex feature extraction from data.

Purpose of the Study:

• To provide an overview of current deep learning-based gait recognition methods.
• To analyze the development and applications of gait recognition in forensics and security.
• To discuss challenges and the effectiveness of deep neural networks in addressing them.

Main Methods:

• Analysis of state-of-the-art deep neural network architectures: Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and attention mechanisms.
• Evaluation of diverse gait recognition models including GaitNet, GA-ICDNet, MSTFE, GaitGraph, and others.
• Assessment of model performance across various conditions, including walking variations, viewing angles, and clothing.

Main Results:

• Deep learning models demonstrate impressive accuracy in capturing unique gait patterns across different conditions.
• GaitNet achieved 99.7% identification accuracy; GA-ICDNet showed a 0.67% equal error rate.
• Performance varied, with GaitGraph and Koopman Operator models showing moderate accuracy, while GCPFP and MFINet faced challenges with specific datasets and conditions.

Conclusions:

• Deep neural networks are highly effective in overcoming gait recognition challenges, significantly enhancing identification accuracy.
• The study highlights the potential of advanced architectures like CNNs, RNNs, and attention mechanisms for robust gait analysis.
• Future research directions are assessed, building upon the current breakthroughs in deep learning-driven gait recognition.