Video technologies are evolving at a steady pace, and that evolution looks set to continue with machine learning and artificial intelligence having such enormous potential to improve so many areas of technology over the coming years.
Exciting video technology developments in recent years include the combined use of cloud computing and video transcoding for live streaming and improved video delivery. In video transcoding, a cloud server receives a video file or stream and creates multiple versions of individual videos, each with a different bitrate and frame size, so that streams and other video content can reach as many end users as possible (read more on video transcoding or see this example of a an actual transcoding service).
According to predictions based on forecasted research by Cisco, video traffic will account for 82 percent of all consumer Internet traffic worldwide by 2021. As video grows in importance on the Internet, expect to see more developments in video technologies.
The prevalence of Big Data, low-cost storage, algorithmic advancements, and elastic compute have all played a part in the machine learning and AI revolution. This article overviews five ways video technologies are can incorporate machine learning and AI.
Five Uses For Machine Learning & AI in Video Technology
Dynamic Advertisements
The continued surge in popularity of video content on the Internet is driving innovations in advertisements. Between 2018 and 2021, the U.S. digital video advertising industry is forecast to increase by double-digit percentages annually, and the value of the market will reach over $22 billion by the end of that term.
The place for AI in video advertisements lies in its potential for delivering dynamic ads based on the geography, language, and demographics of the viewer; ads that are inserted into videos which have already been produced. Video content producers can structure the creation of their videos in such a way that specific locations can be used as placeholders for advertisements. The AI technology dynamically placed products into these placeholders based on a range of factors.
The use of AI in this way ensures not only that content producers are no longer limited to one specific advertiser per video, but that the whole realm of advertising takes on an even more personalized and localized approach.
Color-Match Video Editing
Internet video audiences are demanding by nature ‘there is pressure on video content creators to frequently release new video content at a fast pace so that they are not left behind or forgotten in a sea of new content. However, behind the smooth operation of a popular Youtube channel with lots of fresh content is a lot of hard work in editing videos so they look great. There is a need for new tools that automate some of the more tedious aspects of video editing.
Adobe has led the way in this regard with color-match video editing, which uses Adobe’s Sensei AI platform in the company’s popular video editing app, Adobe Premiere CC. Often, creators use multiple cameras to make their videos; think of a travel video where a drone, mobile phone, and DSLR might be used in a single video. The problem with using multiple cameras is that the colors rarely match, particularly when the cameras aren’t all from the one brand.
The color match feature introduced by Adobe automatically matches colors using AI, significantly reducing what is a normally time-consuming task for video editors.
Video Generation
In what is still a nascent use case, AI technology has advanced in terms of its ability to generate videos. The idea is that by typing out a simple phrase, AI could make a video about the scene you type. The process works with AI and machine learning combining to study many videos and learning to associate different phrases with different motions.
Work on this use of machine learning and AI in videos is still in its infancy, however, and any discussion on the extent to which it could be adopted is just speculation at this point. Some sources have speculated how generative videos could be used to train self-driving cars on how to best operate in a range of rare weather conditions.
More Efficient Policing
Law enforcement agencies often use video cameras to record interactions with the public to improve public safety and perhaps find evidence for previously unproven crimes. In a single year, for example, the LAPD accumulated 33 years worth of footage from its various cameras. The challenge is that a 20-minute video might only contain a few seconds worth of valuable information, and this is where AI comes in.
Axon, the official AI partner for the LAPD, used its AI technologies to find which parts of a recorded video contain valuable information and condense longer videos into short clips of relevant content. Future widespread adoption of AI by police forces could enable much more efficient use of police time.
Sports Analysis and Reports
IBM Watson is a powerful machine learning and AI platform. In 2017, an exciting use case emerged for Watson in powering analytics, real-time match reports, and highlights at the Wimbledon Tennis Championships in London.
Using 22 years of unstructured data, Watson analyzed 53,713,514 tennis data points to date and used machine learning to decipher the most important factors driving a great Wimbledon champion. Factors such as serve effectiveness, stamina, and the ability to return serves were all accounted for in the analysis.
In a related use case, IBM Watson was used at the U.S. Open tennis tournament to immediately create highlight clips of the most compelling content at the end of each match and push these highlights out to social sites to drive more interest in the tournament. Expect to see AI and machine learning innovating more sports in the coming years.
Wrap Up
These exciting use cases provide evidence that video technology is already being shaped by machine learning and AI. As video becomes an even more important content medium, and as companies and individuals use AI more inventively and intelligently, it’s likely to lead to an extraordinary range of possibilities and use cases.