(THIS ARTICLE IS MACHINE TRANSLATED by Google from Norwegian)
Curious spectators gather around an exhibition table in the small military hall of the Oslo Military Society. On the table is a detailed model of Holmenkollbakken and a screen with 3D representation of the jump hill and surrounding area, which the curious can navigate around as a computer game. The model and production of Holmenkollbakken in 3D is based solely on a series of images in 2D. The Defense Research Institute (FFI) has brought together the technologists and potential users in the Armed Forces to discuss what the latest camera and image processing technology can be used for.
"This is a field that is partly about taking the picture, which is about physics, hardware and optics, and partly about how we process the pictures in computers," Torbjørn Skauli, researcher at FFI, explains to Ny Tid. On the camera front, it is especially the development in what the camera can see that is interesting. "The usual cameras we use the most are the ones that capture visible light," says Skauli. "It is a well-developed technology that is also cheap, and we get very high resolution, many pixels and very good sensitivity, which, among other things, in a military context makes it easier to take pictures at night."
However, such cameras can only see light to about one micron, that is, only a small portion of the light spectrum, according to Skauli.
So-called hyperspectral cameras, produced in Norway, can capture far more information. "The human eye has only three primary colors, while a hyperspectral camera can have a hundred or more," Skauli explains. "That way, the rules of the game are changing what we can see with a camera purely physically. That way we can know much more about what kind of material we are seeing. It is a powerful technique for mapping landscapes in a military context, and for distinguishing things that look just like the eye – for example, separating camouflaged things from nature, ”he says. Hyperspectral cameras are also used in waste sorting, in medical imaging, and in the spacecraft that captured images that recently showed traces of liquid water on Mars.
Can monitor an entire city. Developments in cameras that capture only the visible spectrum are also in the process of providing entirely new applications. Gigapixel cameras placed in a plane can capture an entire city in one image, with such a high level of detail that it is possible to zoom in on individuals. "It's a very interesting and not entirely uncontroversial development," comments Skauli. "It's almost like we have a video of an entire city all the time, and that's obviously something we can benefit from in the military. If we manage to process these large streams of images, we may also receive automatic notification of things that are militarily relevant. But it's clear that it involves massive surveillance. "
Gigapixel cameras placed in a plane can capture an entire city in one image, with such a high level of detail that it is possible to zoom in on individuals.
Camera surveillance in the cityscape is not an unknown phenomenon. The prevalence of cameras in banks, in shops, on public transport and around important buildings has accelerated in the last decade. But so far, it has been necessary to gather information from many different sources in order to get as comprehensive and detailed an overview as one can get with one gigapixel image. On the other hand, much more information about individuals can now be obtained than ordinary surveillance cameras. Dagens Næringsliv reported this summer that surveillance cameras for sale for commercial use in Norway today can recognize faces and record information on skin color and gender.
The quality of today's digital cameras has also led to the production of small satellites based on mass-produced components, which are very cheap compared to before. Despite the modest size, these satellites provide many details. "For example, if you look at Skybox Imaging, which was acquired by Google, they offer a ground resolution that's better than one meter, and others offer better than a half-meter resolution of photos and video that one can buy commercially, ”says Skauli. With such a resolution, for example, you can take pictures of things the size of cars. According to Skauli, however, it is limited for how long a satellite can stare at the same location: “These satellites go away very quickly, so video surveillance is not possible at all times. Low-Earth satellites are traveling at about seven kilometers per second, so they disappear quickly over the horizon again. "
Want to learn computers to watch. At least as important as camera technology are the methods developed to analyze the images. “Images contain a lot of data, and the amount of data increases as the resolution of the cameras increases. At the same time, the performance of computers is also constantly increasing, due to the same unstoppable flow of technological breakthroughs, ”says Skauli. For example, by performing complicated computational operations on the images, one can extract information about man-made structures and create 3D models of them – such as the model of Holmenkollbakken. Image processing can also be used to map large areas in a short time, and to navigate without the use of the GPS system, which will be vulnerable in a war between great powers.
So-called deep learning is perhaps one of the things that will change the way we use cameras to a great extent. The method is that the computer learns to recognize things and situations after being given a number of examples explaining what things are. "It's been such a long time that the computer doesn't understand what it's looking at," Skauli explains. "This will still be true for a while, but it is a fairly strong trend towards us being able to use processing to extract constantly new information from images." So the computers have started to make pretty good interpretations of what is in a picture, and can recognize, for example, different types of vehicles or ships that have been informed about before.
The computing power we currently operate is at an incredible thousands of billions of pixel operations per second. "Perhaps the most important consequence of this is that computers can create information products at a fairly high level of real-time abstraction, so that they can begin to take on some of the tasks that previously required a person to follow the picture," Skauli explains. "Automated processing and interpretation of the images means that we are not dependent on the alertness of the person sitting and looking at the screen." In the future, we may think that guards no longer need to monitor surveillance cameras. Instead, the computer gives an alert when it perceives a suspicious situation or person.
New technology, new challenges. Classifying certain behaviors as suspicious is in itself something that raises ethical issues when used in public space. Depending on how the technology is deployed, people with totally legitimate reasons to behave in a way that is flagged by the system may be more suspicious and monitored than before. Processing technology is becoming so cheap and widespread that we can expect private and commercial players to use it more and more. Will the new opportunities lead to the use of cameras in completely different ways and places other than what we are used to?
"Several major cities have seen drones fly over important buildings without knowing who controls them," says Skauli. The presence of drones is an example that is becoming well known in Norway as well. Combined with the processing technology, images from drones can thus be used to extract a detailed model of the buildings in question. From a military perspective, this presents new challenges. According to Skauli, the technology is becoming so cheap and accessible that even non-state actors – such as terrorists – can more easily obtain what was previously reserved for the military in camera technology.
Aarseth is a freelance journalist and regular contributor to Ny Tid.
tori.aarseth@gmail.com.
