STRL is researching automatic video summarization technology to support the production of short videos for online distribution in order to showcase TV programs to viewers. This exhibit shows a system to automatically generate a summary video of a program using scenes selected by image analysis AI.
Using image analysis AI trained to recognize the composition and camera motion of important scenes
The image analysis AI used in this system has learned to recognize the characteristics of composition (appearances of people, objects, scenery, etc.) and camera motion of important scenes that should be included in the summary video. By selecting extracts using this AI, it is possible to generate summary videos with similar quality to those edited by an actual video producer.
Users can easily modify the automatically generated summary video
Short-video producers have various aims or preferences, such as “avoid close-ups of faces,” “keep camera movement slow,” or “keep each cut short.” To respond to these precise requirements, the system includes functions enabling users to easily modify the automatically generated summary video.
Practical use of news program automatic summarization system applying this technology
STRL has developed a “news program automatic summarization system” combining image analysis AI and speech recognition technologies that is now in practical use at news production sites. Summary videos of news programs can be generated and shared on social media immediately after broadcasting of the program has finished.
As well as expanding the use of the news program automatic summarization system, STRL aims to put automatic summarization systems into practical use for other genres of TV programs.