Face Log Creation from Low-Light CCTV Videos

Abstract

In today’s rapidly evolving technological landscape, surveillance systems have become critical for security and operational management. Extracting accurate facial data from low-light CCTV footage remains a significant challenge due to limited visibility. This research presents a comprehensive methodology to address the complexities of face detection, recognition and timestamp extraction in low-light environments. Our approach focuses on creating detailed face logs with in-time and out-time information for each identified individual. The methodology leverages the Enhanced Deep Curve Estimation (EDCE) technique to improve visibility, followed by the Dual Shot Face Detector (DSFD) for precise face detection in enhanced video frames. FaceNet is employed for robust face recognition, while a combination of the Kalman filter and tesseract OCR enables accurate face tracking and timestamp extraction. All extracted data, including facial details and timestamps, are systematically logged into an Excel file for further analysis. The integration of these techniques offers significant advancements in overcoming the challenges of face identification in low-light conditions, presenting a promising solution for enhanced surveillance systems.