Facial Recognition and Image Processing in Security and Surveillance Systems Using Deep Learning Algorithms and Multi-task Cascade Deep Convolutional Neural Networks

This paper provides a significant angle and practical framework for the facial recognition and image procession system in security and surveillance application and used Deep learning Algorithm Multitask Cascade Deep Convolutional Neural Networks (MTCNN). The proposed system will aim to overcome real time difficulties in face detection alignment and recognition especially when occlusion is present, or when the lighting varies, and when there are pose variation. The application of MTCNN framework involves three phases which are P-Net, R-Net, and the O-Net for accurate face detection and a precise predication of facial landmarks that enhances face alignment and therefore feature extraction. The performances of the system is measured using public standard datasets where the system yields 97.8% accuracy with 97.6% of precision and 96.9% of recall. Additionally, the methodology has superior detection reliability under adverse conditions as compared to traditional algorithms such as Viola-Jones and SSD, has an FPS of 25, rendering real-time activity. It is also demonstrated that the proposed system achieves low false positive and false negative rates, especially in lo-light crowded applications, such as modern security and surveillance systems.