Evaluation of Face Detection Algorithms for Real-Time Surveillance Application

Intelligent surveillance technologies have been implemented in different scenarios, requiring robust and efficient face detection and recognition systems capable of operating in real time, often on resource-constrained hardware. In this paper, we propose a CPU-optimized real-time face detection and recognition framework based on the ResNet-10 Single Shot MultiBox Detector (SSD) for face localization and the FaceNet-512 embedding approach for identity verification. In addition, we incorporate a centroid-based tracking method to reduce computational redundancy across consecutive frames. Our experimental results show that the proposed system achieves a mean Average Precision (mAP) of 0.937 and operates effectively in real-time at 42 frames per second on standard CPU hardware, demonstrating strong potential for practical surveillance applications in different scenarios.