| Abstrakt: |
The goal of this work is to use deep learning and machine learning to develop a real-time framework for the identification and recognition of human faces in closed-circuit television (CCTV) images. A typical CCTV system requires constant human monitoring, which is costly and insufficient. The automated facial recognition technology in CCTV footage can help a lot of businesses, including law enforcement, by lowering expenses and requiring less human intervention to identify suspects, missing people, and anyone entering restricted areas. However, there are other issues with image-based recognition, such as scaling, rotation, and fluctuations in light intensity or busy backdrops. This paper suggests using a variety of face recognition and feature extraction techniques to build a human face recognition system based on CCTV photos. Face detection, location, extraction from the captured photos, recognition, and image pre-processing are the steps that make up the proposed system. CCTV is used to obtain images. Convolutional neural networks (CNN) and principal component analysis (PCA) are the two feature extraction technologies used. A comparative analysis is conducted between the K-nearest neighbor (KNN), decision tree, random forest, and CNN algorithms. Recognition is achieved by applying these techniques to a dataset of more than 40K real-time images taken under different settings (e.g., rotation, light intensity, and scaling for simulation and performance evaluation). Ultimately, over 90% accuracy was attained, and the processing time needed to identify faces was minimal. [ABSTRACT FROM AUTHOR] |
