Fujitsu optimizes AI-based video recognition with high-quality compression technology

High-volume, ultra-high-definition video data is compressed to 1/10 the data size of conventional compression technology, for optimized analysis in the cloud, using a new technique.

Fujitsu Laboratories has developed a technology to compress full definition video data and high volume at the minimum size necessary, for video recognition applications of artificial intelligence (AI). This technology can compress video data to only one-tenth the size of data prepared with conventional compression technology, intended for visual confirmation by humans.

In recent years, there has been a sharp increase in demand for AI analysis of video data in various business areas. The expansion of the 5G mobile communications system, in particular, is expected to contribute to an explosive increase in the number of ultra-high-definition video images captured by cameras, as well as many images taken on the street and on production lines.

When developing this new compression technology, Fujitsu focused on an important divergence in the way AI and humans recognize images. That is, AI and humans tend to differ in the areas of the image that are emphasized as important, when recognizing people, animals or objects in video data.

For this reason, the company has developed a technology to Automatically analyze areas under AI values and compress the data to the minimum size that it can recognize. This makes it possible to analyze a large amount of video data without compromising recognition accuracy while significantly reducing data transmission and operational costs. It is also anticipated that the technology will allow users to analyze video information in a more advanced way, combining multiple video data stored in the cloud, sensor data and performance data, such as sales data.

Background and challenges

In recent years, the technology for analyzing images using AI has developed rapidly and is expected to be one of the driving forces of digital transformation in many companies across different sectors. With the arrival of sophisticated 5G mobile services in 2020, the demand for AI analytics will increase further, accompanied by the growing use of cameras 4K and 8K ultra high definition and large amounts of video data for applications including behavioral analytics in manufacturing and retail industries.

Despite this, the processing demands for deep learning techniques used for image analysis present considerable challenges. An effective technique to ensure technological power to tackle these tasks is to co-process with the cloud, but since video data is often resource-intensive, there is a need for high-compression technology that can transmit all video data to the cloud without compromising quality, so that network bandwidth is not overloaded.

The technology

Video compression reduces image quality depending on the compression rate, and if the area the AI ​​focuses on becomes excessively compressed, recognition accuracy decreases. Fujitsu has developed a video compression technology that automatically analyzes the area of ​​an object recognized by AI as analysis material in a 1-frame image of video data, compressing the image to the minimum quality required for recognition of each area. By applying this technology, the size of video data can be significantly reduced compared to conventional compression technologies while maintaining recognition accuracy.

The technology to automatically estimate the compression ratio without affecting the accuracy of AI recognition

The effect of compression-specific image quality degradation on recognition accuracy is analyzed for each area. The compression ratio that does not affect the recognition accuracy is automatically estimated based on the AI ​​recognition results. The degree of importance of the features in the recognition process by the AI ​​is determined for all areas, adding the effects on the recognition results, when the compression ratio of the entire image and also the image quality are changed. The compression rate immediately before the recognition accuracy deteriorates rapidly in each area is estimated as a compression rate that does not affect the recognition accuracy.

It also feeds back the AI ​​results of successive images to increase compression to the maximum that the AI ​​can recognize. By doing so, the technology achieves high image compression, while maintaining the accuracy of AI recognition.

Future plans

The newly developed technology was applied to video footage taken by a 4K camera of multiple workers packing in a factory. It was confirmed that Data size could be reduced to 1/10 of the data size of conventional compression technology, without a deterioration in recognition accuracy. This technology is expected to be used for applications that do not require strict real-time performance, as well as for advanced video data analytics that combine multiple video data stored in the cloud, sensor data, and performance data such as sales data.

Fujitsu Laboratories is evaluating this technology in a variety of cases and is conducting additional research and development to further improve compression performance. The multinational expects to commercialize this technology by the end of fiscal year 2020 and introduce it in a variety of applications for different industries, including its COLOMINA service platform, a Manufacturing Industry solution.