Real-Time Face Detection for Public Safety Monitoring Using Deep Learning and Prompt Engineering

One of the fields that can be mentioned as applied application of the technique is public safety surveillance, where accurate face detection in real-time is needed. The conventional approaches related to surveillance with the use of rule-based approaches or classic machine learning were found ineffective for detecting faces in changing environment, occlusion, and among heterogeneous populations. Therefore, this paper proposes a comprehensive framework to improve the accuracy of face detection using prompt engineering and CNN. The suggested framework implies performing inference, prompt generation, and context-based decision making with the help of large language models (LLMs). As far as results obtained from the experiments are concerned, the accuracy, sensitivity, and frames per second (FPS) for the suggested CNN technique amount to 97.2%, 96.60%, and 63 correspondingly. It means that they proved to be more effective than the measures for conventional techniques (Haar cascade – 80.3% and 45 FPS; HOG+SVM – 80.8% and 28 FPS).