A View Transformation Model Based on Sparse and Redundant Representation for Human Gait Recognition.
Abbas Ghebleh1, Mohsen Ebrahimi Moghaddam1
1Department of Computer Engineering and Science, Shahid Beheshti University, Tehran, Iran.
Journal of Medical Signals and Sensors
|October 16, 2020
Summary
This study enhances human gait recognition by using a novel view transformation model. The method improves identification accuracy, especially when view angles change significantly.
Area of Science:
- Biometrics
- Computer Vision
- Pattern Recognition
Background:
- Human gait is a promising behavioral biometric for identification.
- Gait recognition systems face performance challenges due to variations in viewing angles.
- Addressing view angle variations is crucial for robust gait analysis.
Purpose of the Study:
- To improve the performance of human gait recognition systems.
- To overcome challenges posed by significant changes in view angles.
- To develop a view transformation model for enhanced gait identification.
Main Methods:
- Proposed a view transformation model utilizing sparse and redundant (SR) representation.
- Trained angle-specific dictionaries for SR representation.
- Implemented view transformation by mapping SR representations between dictionaries.
Main Results:
- The proposed method demonstrated satisfactory performance on the CASIA Gait Database (Dataset B).
- The method outperformed existing approaches in most tests, particularly with large view angle changes.
- Achieved higher average recognition rates compared to other methods.
Conclusions:
- The developed view transformation model shows superior performance in gait recognition.
- The method is particularly effective in scenarios with substantial variations in viewing angles.
- Outperforms state-of-the-art methods, offering a more robust gait identification solution.
Keywords:
Biometricsgait analysishuman identificationsparse and redundant representationview transformation modelview-invariant
