Gait Recognition Using Deep Neural Networks

Please enable javascript in your browser.

Karimi, Ali | 2025

0 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 58192 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Kasaei, Shohreh
Abstract:
Gait recognition is a biometric method that aims to extract an individual's walking patterns and identify their identities. This recognition method has gained significant attention due to its ability to operate over long distances without requiring direct interaction. Additionally, these extracted walking patterns can be beneficial in the healthcare field, particularly for early diagnosis of certain diseases. With the advancement of deep neural networks, remarkable progress has been achieved in this area. However, numerous challenges remain, including variations in clothing, occlusions, and recognition in practical scenarios. For recognition models to make accurate decisions in real-world scenarios, they must effectively extract relevant features from RGB data while being lightweight in design. This research introduces KEGait, a lightweight end-to-end architecture trained using a knowledge distillation mechanism. The KEGait model has demonstrated promising results on one of the latest datasets, CCPG, achieving competitive performance in evaluation metrics such as Rank-1, mAP, and mINP
Keywords:
Biometrics ; Surveillance System ; Deep Neural Networks ; Gait Analysis ; Knowledge Distillation ; End-to-End Architecture