A Smart Surveillance System for Uncooperative Gait Recognition Using Cycle Consistent Generative Adversarial Networks...

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• Gait recognition is a behavioral biometric system for identifying individuals by their walking patterns.
• Existing gait recognition systems struggle with accuracy in uncooperative environments due to varying walking conditions.
• Smart surveillance demands high accuracy, even with uncooperative subjects.

Purpose of the Study:

• To develop a deep learning method for robust gait recognition in uncooperative environments.
• To address the limitation of gait recognition systems facing diverse and unknown walking conditions.
• To enhance the accuracy of person identification using gait analysis under dynamic conditions.

Main Methods:

• A deep learning approach combining Convolutional Neural Networks (CNNs) and cycle-consistent Generative Adversarial Networks (GANs).
• The method translates gait energy images (GEIs) from various covariate factors to a normal GEI using unsupervised learning.
• Cycle-consistent GANs with a Cycle Loss function identify individual pairs and generate normalized GEIs for CNN-based person identification.

Main Results:

• The proposed system achieved excellent results on the publicly available CASIA-B dataset.
• Demonstrated successful gait recognition even with disturbed and varied walking conditions.
• The translation of disturbed GEIs to normal GEIs significantly improved recognition accuracy.

Conclusions:

• The proposed deep learning method effectively handles uncooperative environments in gait recognition.
• This system offers a viable solution for security applications in sensitive areas.
• The approach enhances the reliability of behavioral biometric systems for person identification.