Research On The Segmentation Of Instructional Videos Based On Visual Information
Aowei Hu, Kun Yu, Zheng Xiao, Wenxin Hu · 2024
In the context of online education, a vast number of instructional videos have emerged. Precise segmentation of instructional videos based on knowledge points is of significant importance for tasks such as knowledge graph construction and adaptive learning recommendations.We propose a target detection model based on layout analysis to extract valuable structured information from video frames. Simultaneously, by integrating text recognition techniques and natural language processing methods, the intelligent slicing of individual knowledge point videos within instructional videos is achieved. In our work, layout analysis is considered a target detection task, and a novel metadata set for Instructional Video Frames Layout Analysis(IVFLA), similar to Document Layout Analysis(DLA), is innovatively introduced to assist in locating and classifying elements on video frame images.Experimental results demonstrate that the proposed method effectively detects nine classes of elements in images, with a precision metric reaching 0.98665 in training and 0.887 in validation, exceeding other classic models. Our model has been tested to exhibit predominant accuracy and efficiency in the precise segmentation of knowledge point videos.