Deep Learning Based Text Detection Model for Lecture Video Analysis: Impact of Covid 19 in Education Sector

The global education system is impacted by the corona virus. Therefore, online learning is a way to keep the educational system going. Learners who are suffering from health issues cannot sit for long time to watch lecture videos. It is not possible to navigate directly to the precise needed point of topic in the video when students are just interested in seeing the desired chunk of topics from the lengthy lecture video. Therefore, we are presenting a study on lecture video indexing to quickly and non-linearly access the topic of interest in a long lecture video, which can help learners who cannot go to the offline classes and who are suffering from health issues. Key frame representation is an effective method for getting accurate video content. Thus, a faster R-CNN-based improved algorithm is proposed to detect text from the key-frames for index points generation. The experimental findings demonstrate that the accuracy of keyframe extraction of proposed method is 90%, and the optimized Faster R-CNN algorithm paradigm significantly increases the detection accuracy to 93.4%, which is the best compared to other algorithms and minimizes the skipped detection performance.