Fujitsu Labs Develops High-Comp Tech  to Boost AI-based Automatic Video Surveillance

Fujitsu Labs  Achieves overwhelming high compression using AI-based Automatic Video Surveillance while maintaining image recognition performance by deep learning

Fujitsu Laboratories Ltd. is a video compression that can recognize the contents of video with high accuracy even from the ultra-highly compressed video. Automatic Video Surveillance technology is one of the most powerful solutions in the field of surveillance and tech-assisted security.

Details of this technology will be announced at the “Picture Coding Symposium of Japan 2020”.

In recent years, due to the dramatic improvement in the performance of image recognition through deep learning, image recognition AI solutions that extract information from a video have attracted attention, and the development of video compression technology specialized in AI recognition has become active. In particular, when automating monitoring and confirmation work with AI, a technology that compresses and transmits only the deep features required for image recognition by AI without restoring the image is drawing attention.

This time, we applied our original high-dimensional data analysis technology “Deep Twin” to the compression of deep feature quantities, and applied the conventional H.265 without degrading the image recognition accuracy. We have developed a video compression technology that can achieve a compression rate of 100 times or more compared to a general method that combines video compression with AI. By utilizing this technology, we will suppress the increase in the amount of video transmission data, which is becoming larger due to the spread of image recognition AI solutions, and realize the efficient use of limited communication resources, making the world more sustainable.