WHEN THE US Air Force deployed Gorgon Stare, a drone video system that consists of 368 cameras covering nearly 40 square miles at a time, in 2011, an official declared, “we can see everything.” The technology, named after snake-haired mythological creatures whose gazes turn people to stone, can surveil an area for hours at a time, take composite images of 1.8 billion pixels each, and create several terabytes of data every minute.
According to its latest 25-year unmanned systems roadmap, the Pentagon operates more than 11,000 drones, and the vast majority of them conduct video surveillance. In 2011 alone, the US Air Force amassed over 325,000 hours of drone video—that’s about 37 years of video gathered by one military service in one calendar year, and that was six years ago.
While drone strikes have altered armed conflict over the last several years, the use of unmanned aircraft systems for surveillance and intelligence may turn out to be a more revolutionary development. But this is contingent on our ability to review all of the collected video, analyze it, and derive insights and intelligence. And frankly, there is just too much of it to reasonably review.
What we’ve learned is that drones can see everything; we can’t. The imagery and video are coming in so quickly and in such high volume that it often overwhelms decision-makers. This is exactly what happened one winter day in 2010 when a senior commander cited “information overload” as the cause for a drone strike that resulted in the deaths of 23 Afghan civilians.
With demand for more drone operations continuing unabated, the military should devise ways to review every second of drone video surveillance and manage the wealth of information that results. The answer to this quandary is a mix of emerging artificial intelligence, analytics, and compression technologies that would automate the review and initial analysis of drone video. In other words, the military should teach its machines how to watch TV.
The benefits of this approach are immediately obvious. By ceding video analysis to machines, the military will be able to leverage all the collected data instead of the extremely small percentage currently examined—a Defense department official tells us that 99 percent of all drone video has not been reviewed. If the Pentagon can tap into that trove of video, it could obtain a depth of knowledge about its adversaries and areas of operations that would be unrivaled by any nation or non-state organization.
This could alter the character of conflict in the same way that precision-guided munitions did decades ago. Moreover, more thorough analysis of drone footage could reduce the deaths of innocent civilians in conflict zones, speed aid and recovery efforts in a humanitarian crisis or in response to a natural disaster, and possibly take out enemy combatants before they can carry out attacks.
The concept is fairly straightforward: First, artificial intelligence and video analytics review stored drone video to create a history of activity and characterize what’s happened. Though the military employs trained intelligence analysts to spot anomalies or events of interests, there is simply too much collection to review. Further, 60 percent of drone video is benign and holds little to no value, resulting in analysts staring at a screen waiting for something to move—“it’s just a total waste of manpower,” retired Marine Corps General James Cartwright has said. AI and analytics can take over this task and establish a baseline of activity, helping military leaders identify patterns and behaviors associated with everyday events, indications of an imminent attack, and everything in between.
Then, AI could be used when drones are in flight to compare the freshly collected video against the patterns identified from the stored video. For example, video analytics may reveal that every time a small convoy of white pickup trucks traverses a particular segment of deserted terrain, an attack occurs in a nearby city within a couple days. Today, when a drone picks up this convoy, it’s hard to know if this is an event of interest or just harmless trucks crossing the land. But once viewed against historical patterns, that convoy might suddenly become a high-interest occurrence that military analysts can focus in on and monitor—or even take action before a terrorist attack occurs.
But high-definition video files are huge. So increasingly large amounts of expensive bandwidth are required to view video real-time for every combat and surveillance mission. A 2015 report from the US Government Accountability Office notes that technical configurations and the high-priority of certain drone missions require leasing commercial satellites for video transmission, contributing to a $1 billion annual bill, in addition to the costs associated with maintaining military-specific satellite communications. As such, the military needs onboard video compression technologies so that transmission of the higher-definition video required to facilitate AI and analytics examinations is practical, less costly, and more efficient.
The combination of artificial intelligence, video analytics, and compression technologies will lead to an exponential increase in the utility and value of military drones. And given that the current Defense bill sets aside nearly $7 billion for drone-related research, procurement, and construction with $2.6 billion going toward existing unmanned aircraft, extracting the maximum value from drones is good for national security as well as for the Department’s budget.
Of course, this recipe for revolutionary insights and action is easier said than done. The technology and algorithms required to make this reality are far from perfect. For example, in a recent object recognition test the Defense Department conducted, its experimental drone had trouble distinguishing a minaret from an armed man. And experts note that video analytics continues to face issues with camera types, lens placement, and vision and machine learning algorithms. But the gains from figuring out these challenges far outweigh the time and investment required to realize this vision.
The Defense department recognizes the importance of solving these issues, so it has undertaken a number of initiatives to integrate artificial intelligence and machine learning into surveillance and combat. One such effort is Project Maven, which seeks to automate basic labeling and analysis associated with full-motion video surveillance. The Navy has also created a digital warfare office to bring the power of data analytics to improve a range of missions. Of note, these endeavors require the use of commercial technologies and best practices, so any solution will be the result of deep collaboration with the private sector.
Much of the conversation concerning AI and drones is dominated by the question of autonomous machines launching missile strikes, though Vice Chairman of the Joint Chiefs of Staff General Paul Selva has indicated the US was a decade away from having such capability and had no plans on building it. But the more pressing question may be whether the military can afford to operate without the fullest insights from deep analysis of drone collection and the advantages of better and quicker decision-making processes.
Not only does this reduce civilian deaths, collateral damage, and response times to crises, but it also gives the military a decisive advantage over its adversaries.
As drones become more and more integral to military operations, the video from their cameras will become increasingly important. Ensuring that collection isn’t wasted requires a technological feat that’s difficult but well within reach. Once achieved, it has the potential to transform military operations.