Real-Time Face Detection and Recognition on Raspberry Pi using LBP and Deep Learning
Venkata Kranthi B, Surekha Borra · 2021
Advancement in low power embedded systems in recent years help us in building portable real-time algorithms and applications. These embedded systems can be integrated into the Cameras used in Surveillance Systems to provide additional mid-level computer vision tasks like face detection and face recognition in person identification applications. This paper proposes face detection using Local Binary Patterns (LBP) and Haar cascades-based face recognition using Convolutional Neural Networks (CNN) derived from Lenet architecture. A database is created covering all challenges involved in face identification like illumination, orientation, expressions, disguise, and age factors. Two CNN architectures are proposed and compared for face recognition. The tasks are performed on Raspberry Pi in real-time, and analysis has been carried out on how well LBP and Haar cascades work in terms of accuracy and Frames per second (FPS) in real-time. The realtime results achieved are acceptable with the frame rate of 7.04 FPS with accuracy above 94%. Frame reading and frame processing are handled in separate threads on CPU and frame skipping, while detection and recognition have improved the frame rates significantly.