Indoor drones able to fly autonomously with image processing algorithm
Image credit: Dreamstime
Researchers in Japan have developed a single-camera machine vision algorithm that allows lightweight hovering robots to guide themselves by identifying and interpreting reference points on a tiled floor.
The team behind the project, led by Chinthaka Premachandra, an associate professor in the Department of Electronic Engineering at Shibaura Institute of Technology in Tokyo, said the technology “opens the door to a new breed of functional, low-cost drones with potentially wide-ranging uses”.
As GPS signals are too weak to penetrate most structures, indoor drones must rely on environmental cues which are typically visual, Premachandra said. Also, a drone designed for indoor use is likely to be smaller and lighter than an outdoor drone.
“We considered different hardware options, including laser rangefinders,” he said, “but rangefinders are too heavy and infrared and ultrasonic sensors suffer from low precision. That led us to use a camera as the robot’s visual sensor. If you think of the camera in your cell phone, that gives you an idea of just how small and light they can be.”
The research team designed the guidance algorithm to be as simple as possible, allowing the use of a small and inexpensive microprocessor. The team chose the Raspberry Pi 3, an open-source computing platform that weighs approximately 45g.
The prototype the team developed had a single downward-facing camera with intentionally low resolution – only 80 by 80 pixels. “Our robot only needed to distinguish its direction of motion and identify corners. From there, our algorithm allows it to extrapolate its position in the room, helping it avoid contacting the walls,” Premachandra said.
According to the team, the program worked by taking each 80 x 80 image through a series of simple processing steps which resulted in a black and white grid. This made it easier to quickly identify motion along the X and Y planes. Premachandra stressed, however, that this guidance method is limited because it relies on a room with a tile floor and predictable patterns.
He added the next steps in research into lightweight, autonomous-hovering indoor robots might include adapting the technology for infrared cameras so they could function in the dark, as well as adding a second camera so the robot could visually determine both its X, Y position and its altitude in the room.
“There are many potential applications,” Premachandra said. “Hovering indoor robots may be useful in warehouses, distribution centres and industrial applications to remotely monitor safety.”
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.