Auto Identifying Key Information in Video Lectures to Generate a Navigation Structure

Online learning platforms are crucial today to ensure uninterrupted teaching and learning. Lecturers upload recorded lectures on these platforms, aiding students having connectivity issues and supporting distance education. However, these lecture recordings are often too lengthy to efficiently navigate to the most attentional content, which includes essential information provided in the lecture. This study proposes a new method to navigate lengthy videos by automatically identifying the most attentional content in video lectures. It would be a better alternative to the existing findings, such as text search in the audio transcript or manual video pausing. Three sub-questions drove the research: what are lecturers' most common techniques to emphasise the most attentional content, how can these techniques be identified in slides, and how can a navigational structure be created to access the most attentional content in video lectures? The research employed a constructive research approach. It started with a questionnaire to uncover lecturers' common techniques, such as text highlighting, underlining, and writing on the slides to denote the most attentional content. The study included a comparative analysis of existing image processing techniques such as Contour Detection, Canny Edge Detection, and Probabilistic Hough Line Transformation, and a machine learning object detection model named EfficientDet DO 512x512 to identify the most attentional content in video frames. The findings revealed that the object detection model was more effective in identifying the most attentional content. Finally, the research employed a strategy involving video frame capturing, slide transition detection, tagging video frames with timestamps, and relevant transcribed words to generate the navigation structure. The findings demonstrated that the navigation structure could effectively locate slides containing the most attentional content with high precision values. Hence, this research can probably aid in the deeper analysis of video frames and enhance student learning from video lectures in online courses.