- Published June 2018
Computer vision, image processing and drones go hand in hand as high-resolution cameras become smaller, lighter and smarter. We look at what influence AI is having on the products and solutions reaching the UAV market.
Inspection drones aren’t particularly new, but the way in which these miniature flying machines can survey, record, report and even fix issues is changing. Now able to fly for tens of miles along power cables or pipelines, sifting through the amassed data would be a full-time job for a team of humans. Thermal, LiDAR and 3D imaging camera technology has been successfully miniaturized and is being mounted on smaller drones, enabling extended battery life and greater reach. AI software is now facilitating drones to “bypass” flawless pipes, towers and cables and focus on weaknesses and defects so that only critical images are recorded for review. Avitas Systems claim to have nailed the “first inspection solution offering enhanced, robotic-based autonomous inspection, advanced predictive analytics, digital inspection data warehousing, and intelligent inspection planning recommendations”. This would open the door to not only all-seeing and all-knowing drones, but devices that predict the future as well!
Neurala’s Brains for Bots SDK aims to make cameras and inspection drones more intelligent and interactive: “The SDK transforms any type of drone device into an intelligent ‘situational partner’ that can perform operational assignments both on- and offline”. Originally developed for NASA’s planetary exploration, Neurala’s deep learning software can now be supported on-board, meaning that a drone can make its own decisions about obstacle avoidance and item recognition, and even carry out remedial action without needing to communicate with a PC or supervisor. Optimized for mobile deployment, the software can be trained using just an estimated 20% of the traditional number of images. Their search technology has been developed specifically to examine shifting environments for moving targets in real-time, enabling the discovery of a mobile needle in a traveling haystack under critical time restraints.
Drones with the ability to survey huge expanses of agricultural land and autonomously identify areas which require, for example, additional pesticide care, are saving millions of dollars in over-spraying, and, of course, avoiding unnecessary environmental damage from excess chemicals. The importance of drones in agriculture was highlighted by the 2017 NVIDIA Inception Award which recognized the work of Gamaya in combining big data processing via AI and hyperspectral imaging to manage planting gaps, weed detection, nutrient levels and soil erosion while predicting crop yield. This particular application was developed for use in Brazil but when your farm spans thousands of hectares of the world’s most remote terrain, these little tools could be a valuable investment.
Zipline is a notable example of drone technology for the greater good, delivering blood and vaccines to isolated areas of Rwanda. Combine the sense-and-avoid software coming onto the market, and this sector can benefit hugely from flying AI. And Zipline aren’t alone, VillageReach, Flirtey, Vayu Drones and (naturally) Google’s Project Wing are just a few other organizations investing research and capital into intelligent drones – Ehang are even working on developing a drone which can carry a person with the aim of expediting organ donation. While stringent flight regulations in many first-world countries may pose a barrier to implementation, less developed nations are an accessible and practical testing ground.
AI in defense is a contentious area (see our AI post from 2017), but huge profits for successful developers continue to make it one of the driving forces in AI research. Systems Technology Inc (STI) is partnering with other US military contractors to develop a hardware and software solution for UAVs which can be launched from the deck of a warship and controlled by hand gestures. Deck Intelligent Aircraft Body Language Observer, or DIABLO, aims to mitigate the noise and constant distractions within this environment and use machine learning to teach drones to recognize the movements of deck handlers, leading to more efficient launches and landings. Reports suggest that research has overcome the common challenges posed by in-the-field computer vision applications including low light, sun glare, temporal resolution and scene clutter. However, the use of Google’s TensorFlow AI systems to analyze the video footage of military drones has caused consternation within the company, and in May this year accounts emerged of several employees resigning, citing concerns over Google’s collaboration with the Pentagon.
Recent developments in UAV technology are bringing the fantasies of aircraft engineers to life, to the benefit of a range of sectors. The devices are capable of capturing massive amounts of visual data, and now technology exists to analyze this and extract relevant information in previously inconceivable timeframes. Completely autonomous drones are the latest aspiration, and it really won’t be long before intelligent flying robots are an expectation rather than a dream in many areas of life.