Human Face Detection and Tracking Using RetinaFace Network for Surveillance Systems

Face detection is the main component in the development of CCTV-based Intelligent Surveillance System. Face Detection is used to identify a person when a suspicious event occurs. Therefore, the face detection module must be reliable and fast in analyzing every frame produced by CCTV. RetinaFace is a deep learning-based face detection method that produces very high accuracy. However, RetinaFace cannot be fully implemented directly on the ISS due to limitations in detecting faces in environments with illumination changes. Thus, in this paper, we propose to utilize the detection-based tracking to improve the detection results of RetinaFace as post-processing stages. The tracking system successfully increases the recall score of the detection of faces on recordings with 25 FPS by 4.47%.