Fast and accurate face recognition system using MORSCMs-LBP on embedded circuits

Area of Science:

• Computer Science
• Biometrics
• Artificial Intelligence

Background:

• COVID-19 pandemic necessitates contactless technologies.
• Traditional biometric systems (fingerprint, PIN) pose infection risks.
• Advancements in embedded systems like Raspberry Pi enable sophisticated applications.

Purpose of the Study:

• To develop an efficient, contactless facial recognition system.
• To integrate local and global feature extraction for improved accuracy.
• To implement the system on a low-power embedded device.

Main Methods:

• Developed MORSCMs-LBP approach combining Local Binary Pattern (LBP) and radial substituted Chebyshev moments (MORSCMs).
• Implemented the system on a Raspberry Pi 4 using C++ and OpenCV.
• Extracted local and global features to create a unified feature vector.

Main Results:

• Achieved high accuracy on benchmark datasets: 99.03% (face95), 99.44% (face96), and 100% (grimace).
• Demonstrated superior performance compared to other recent facial recognition methods.
• Successfully implemented on resource-constrained Raspberry Pi hardware.

Conclusions:

• The MORSCMs-LBP approach offers an effective and efficient solution for contactless facial recognition.
• This method provides a viable alternative to traditional, touch-based biometric systems.
• The system's performance on embedded hardware highlights its practical applicability.