Imagine being a plant manager and you’re looking for a more secure and effective way to inspect your equipment. It could be smokestacks, pipelines, heavy machinery, conveyor belts, or even a complex system with many moving parts that span a great distance from one end of a compound to another. Currently, most companies are still performing manual inspections or, if they are using drones, they are inspecting equipment and systems using human-operated drones. Percepto is a company that wants to change that by allowing industrial drones to fly autonomously and inspect equipment around the clock with real-time reporting.
Percepto got its start in late 2014 when Dor Abuhasira, Raviv Raz, and Sagi Blonder put their combined expertise together to solve a problem. CEO Abuhasira and Chief Technical Officer Blonder both have master of engineering degrees in computer vision from Tel Aviv University and Chief Operating Office Raz holds a bachelor of engineeing from Ben-Gurion University. Being one of two companies working on the solution, they are on the cutting edge of machine vision technology.
Percepto’s Underlying Vision and Sensor Technology
There are three core components to Percepto’s technology – computer vision, sensor fusion, and machine learning.
“We understood two things very fast,” said Ariel Avitan, Percepto’s chief commercial officer. “There is no computing power on drones to run computer vision applications. Secondly, the business matrix of the world will be required to accept computer vision applications because if you buy a thousand dollar drone with vision technology on it, it won’t be affordable, and there’s a big push to move to commercial applications.”
In other words, computer vision technology needs to solve a practical vision problem for companies that use it in order for it to make good business sense for investment. It must be affordable and save the company money. Percepto’s core business is providing the technology to make 24/7 autonomous drone inspections possible.
The core component Percepto’s technology is a Linux-based computer branded PerceptoCore. Weighing less than 50 grams, the computer enables equipped drones to perform computer vision computations on the fly.
PowerCore is built on Ubuntu with a NVIDIA TK1 processor. Although, Avitan said the company is expected to migrate to a TX1 at first opportunity. Further, PowerCore is equipped with three sensors, a 3-axis gyroscope, a 3-axis accelerometer, and a 3-axis magnetometer.
“The GPU enables real-time vision on the drone,” Avitan said.
3 Key Ways Percepto Helps Its Clients With CV and Machine Learning Technology
In addition to PerceptoCore, the company provides a base station for drones that allows them to return to a charging station without the need for human touch. By returning to its base (1.5 meters wide and weighing 200 kilograms) to charge and shelter after each mission, a drone can cycle autonomously. This alone can save a company money in the man-hours it takes to operate their drones.
PerceptoCore allows companies to bypass drone controllers. Equipped with day and night cameras, information collected can be processed in real-time and managed continuously through a management console in the cloud.
Companies that want to develop their own applications can purchase a development kit. Percepto sells its supercomputer to big companies that have programming powers and that want to development their own applications. They can do that on PerceptoCore and install it on their own drones.
Looking at the Future of Drones Empowered by Computer Vision
“There are a lot of companies that focus on vision only,” Avitan said. “They create vision applications that enable you to detect an object, but they don’t play with other features. They can’t recreate the code to run in real-time.”
That’s where Percepto has the edge. There’s only one other company pursuing the same vision and they aren’t any closer to penetrating the market than Percepto is, which means this Israeli company is just out of the starting gate to a race for dominance in a field that isn’t on anyone else’s radar.
“This isn’t an easy place to play,” Avitan said. “With neural networks and deep learning, we have the ability to create next generation applications that will be tailored more specifically to commercial needs. More companies will use these applications in a secure manner, but it’s a very high barrier. It’s not easy to find people who understand neural networks and know how to use them.”
It’s also hard to tailor them to specific applications, but Avitan said it will be much easier in five years.
“You’ll see drones flying and conducting inspections, not based on mechanical or GPS functions, but based on vision technology.”
What Companies Look For in Computer Vision-Drone Technology
There are three primary sets of values to be acquired from continuous autonomous drone inspections in the commercial sector. First, security.
Autonomous drones can fly day and night. With computer vision technology, they can identify humans who are where they’re not supposed to be, cars that have entered a restricted area, and other threats such as misplaced packages, unidentifiable cargo, and even other drones such as those from competing companies, crime syndicates scoping out opportunities, or terrorist organizations.
“We can enable drones to operate without a human operator,” Avitan said. “You can click on a map and have the drone fly to that area, open its camera, and identify a threat. There’s no need to know how to fly a drone, which is key to operating in an industrial environment.”
Then there’s safety.
“This isn’t about vision technology, but more about sensors,” Avitan said. Sensors allow drones to detect leaks in gas pipelines, for instance.
Finally, companies need to inspect their systems. In this case, it’s as much about the radar as about the camera. Drones can identify specific objects and inspect high voltage towers, chimneys, photovoltaic power systems, wind turbines, solar park panels, warehouses, and more.
“The idea is to get to the place where drones inspect objects quickly,” Avitan said. “They can detect object deterioration and allow companies to take control of the maintenance cycle, reduce downtime, and not be surprised when something doesn’t work as it should.”
In other words, fully autonomous drones equipped with computer vision and machine learning applications allow companies to provide around-the-clock preventive maintenance by keeping a constant eye on critical systems. The drone can learn what normal operations look like and detect early signs of a leak, mechanical or material deterioration, and object threats in real-time without the need for manual human assistance.
Avitan said Percepto is currently in “heavy pilots with two very large clients,” a large data company and “the second largest oil and gas company.”