Video Segmentation and Retrieval Based on Image-Driven Person Recognition for Surveillance
Aniket Dhage, Dipali Gangarde, Mrinmayee Deshpande, Harita Joshi, Anuradha Yenkikar · 2025
The significant increase in the use of surveillance systems for public safety resulted in large volumes of video footage and made manual review inefficient and often impractical. Law enforcement agencies are challenged by the effort to identify suspects or missing persons within hours of video footage. It is also a rather time-consuming and laborious task with a lot of scope for error, which can delay investigations or allow criminals to get away with it. The challenges cited above are why the paper derives an automated video-based people-of-interest identification system using advanced image-driven person recognition techniques. Deep learning models, combining with YOLOv10 for efficient face detection, OpenCV for accurate face recognition and mechanisms for video segmentation and retrieval were used. This system processes video one frame at a time, detecting and matching faces against a predefined target image. At the moment that the system identifies the target person, it also retrieves and stitches together relevant video clips, thus conserving much of the time and effort invested in manual video review. This solution improves the speed and accuracy of person identification in large-scale video data, making it highly valuable for law enforcement and security applications. By automating the video review process, the proposed system enhances operational efficiency, reduces human error, and accelerates the decision-making process, contributing to more effective and timely investigations.