User Object of Interest based Video Clip Extraction Using Pretrained YOLOv7

Video clip extraction is creating shorter, focused video segments by identifying and extracting frames containing specific objects of interest. This targeted extraction enables users to access relevant video content without watching entire recordings, proving essential in surveillance, content management, and educational applications. Our proposed object-based video clip extraction approach leverages the YOLOv7 object detection model, combining robust detection capabilities with efficient processing. In our method, we implement frame-by-frame analysis to identify instances of user-specified objects of interest (UOoI). The detected frames undergo temporal analysis to maintain sequence continuity and eliminate isolated detections. We employ a comprehensive object dictionary that supports diverse detection scenarios, from ordinary objects to specific items of interest. Our graphical user interface (GUI) facilitates seamless interaction, allowing users to select input videos and specify target objects for extraction. The framework processes these selections through the YOLOv7 architecture, demonstrating superior detection speed and accuracy compared to previous models. The system generates focused video clips containing the specified objects while preserving the temporal coherence of the original footage. We evaluated our method using the established SumMe dataset and a newly annotated dataset of 21 videos, achieving F1 scores of 73.95% and 69.74%, respectively, with 89.75% and 84.00% accuracy. The experimental results demonstrate that our proposed method produces more precise and relevant video clips than related approaches. This framework represents a significant advancement in automated video content extraction, offering a practical solution for managing and analyzing large-scale video data.

• Supervision