作者: Bhargava Urala Kota , Kenny Davila , Alexander Stone , Srirangaraj Setlur , Venu Govindaraju
DOI: 10.1109/ICFHR-2018.2018.00013
关键词:
摘要: Online lecture videos are a valuable resource for students across the world. The ability to find based on their content could make them even more useful. Methods automatic extraction of this reduce amount manual effort required indexing and retrieval such possible. We adapt deep learning method scene text detection, purpose detection handwritten text, math expressions sketches in videos. detect elements whiteboard generate summary all over time lecture, while also dealing with occluded due motion lecturer. train, test publicly available AccessMath video dataset evaluate our framework basis number frames, as well recall precision set found that increases state-of-the-art there is potential increase well. have added existing ground truth by providing timestamp-based, semantically meaningful bounding box annotations content, which has been released.