Bookmark and Share

Edge Intelligence in the Computer Vision Market

This market research report was originally published at Tractica's website. It is reprinted here with the permission of Tractica.

Smart cameras have been around for some time. In fact, Wikipedia gives an example of a smart camera dating back to 1985. Of course, smart cameras did very simple things back then and the level of intelligence has increased significantly over time. Today, intelligent cameras are driven by computer vision technology.

In the post-9/11 world, everyone and everything is being watched. We are living in a world of cameras.  Frost & Sullivan estimates that, as of 2014, there were 125 surveillance cameras in the United States per 1,000 people. This adds up to more than 40 million cameras in the U.S. alone.  Add the cameras worldwide and now we are looking at many millions of installed cameras. These cameras are tracking people on the street, as well as at airports, train stations, and retail stores – pretty much every public location. They are constantly generating video feeds. Today, these cameras mostly act like dumb terminals in the sense that they take the data and transmit it to a server. In some cases, the software running on the server searches the data for image recognition and takes further action.  In other cases, the data is simply stored on the server for a period of time that varies depending on the application.

The amount of data generated by these cameras is increasing dramatically with each passing year. VGA resolution was the standard in the past, most of the current-generation cameras utilize high-definition (HD) resolution, and pretty soon 4K will be commonplace. This amount of data adds significant capacity requirements not only for networks but also for storage and processing systems. These factors are some of the primary drivers for intelligent cameras. An intelligent camera decides, in real time, which data should be discarded and which data should be sent to the server as it processes and analyzes the video feed. An intelligent camera embeds a computer vision chip in a camera that essentially performs the job of a person watching the video and flags items of interest. In essence, the intelligence is transferred to the edge.

Applications for intelligent cameras are not limited to surveillance. For instance, they could be used in retail environments to track customer behavior without any need for network connectivity. Such cameras can also be used for automotive, home security, and other applications. A computer vision-enabled camera can also enhance images in low-light conditions, as well as removing fog and noise to further improve video quality.

Several companies have recently announced plans for intelligent cameras. FLIR Systems, in collaboration with Movidius, recently announced an enhancement of its thermal imaging camera to include intelligent functionality. Simplicam,  a security camera startup with facial recognition capabilities, is already shipping on Amazon. CEVA recently announced that it has licensed its intellectual property (IP) to a Taiwanese system-on-a-chip (SoC) company called iCatchTechnology to enhance its smart camera technology.  Another startup called Eutecus is focusing on developing IP for edge intelligent devices.

The list could go on and on, but the fact that one can easily embed a computer vision-enabled chip in a camera opens up the field for countless applications. The market for such cameras could easily reach into billions of dollars over the next several years. For example, if we were to consider that in the near future only 1% of security cameras will have intelligence, we have more than 400,000 cameras as a potential market in the U.S. alone. The number goes even higher when we consider the entire world. This quick estimate does not even consider other application markets that would be enabled by smart cameras.

The requirements for such devices will vary and will be dictated by the nature of the applications. For instance, an intelligent camera targeted at the home security market will probably need a standard facial recognition algorithm without the need for 100% accuracy. On the other hand, a security camera at the airport that is capable of tracking a fugitive will likely need sophisticated algorithms to minimize the risk of errors. This will open up the market for chip companies with different performance and power characteristics, IP developers, and device manufacturers.

There is no doubt that computer vision is enabling many interesting applications in everyday life, and edge intelligence is shaping up to be yet another high-profile application.

By Anand Joshi
Principal Analyst
Tractica