VideoCignium addresses the persistent challenge in urban security and forensic analysis: the manual review of extensive surveillance footage, which is labor-intensive and prone to human error. Although advanced AI models exist, their practical integration into accessible tools for non-technical forensic analysts remains limited. This work introduces an open-source desktop application that automates the analysis of security camera recordings by combining computer vision and artificial intelligence. The system detects motion in user-defined regions of interest, classifies objects using YOLO models, and extracts timestamps via OCR, all within an Electron-based interface that enables efficient sequential processing of large video volumes. Validation on high-traffic urban datasets shows a recall of 99.59% and a precision of 94.23% in motion detection (F1-score = 96.83%), with 72.35% accuracy in object classification, resulting in a review time reduction of up to 40x compared to manual methods. The software integrates temporal metadata extraction and structured reporting, providing a robust and open framework for urban security and data-driven policy making.
Natera et al. (Tue,) studied this question.