Lecture Video Automatic Summarization System Based on DBNet and Kalman Filtering

Video summarization for educational scenarios aims to extract and locate the most meaningful frames from original video based on main contents of lecture video. Aiming at defect existing computer vision-based methods that tend target specific scenes, a method content detection tracking is proposed. Firstly, DBNet introduced detect such as text mathematical formulas in static these videos, which combined with convolutional block attention module (CBAM) improve precision. Then, frame-by-frame data association instances performed using Kalman filtering, Hungarian algorithm, appearance feature vectors build tracker. Finally, segmentation key frame location extraction are according instance lifelines deletion events constructed by tracker, extracted groups used final summary result. Experimenting variety video, average precision 89.1%; recall results 92.1%.