GIST Develops Sensor Calibration Technology for Ultralightweight Drones

(From left) Minji Kim, student of Mechanical Robotics Engineering, and Professor Pyojin Kim.

Gwangju Institute of Science and Technology (GIST) announced on September 17 that the research team led by Professor Kim Pyojin from the Department of Mechanical Robotics Engineering has developed a new sensor calibration technology called 'SPLiCE (Single-Point LiDAR and Camera Calibration & Estimation),' which can be used even in ultralightweight drones.

This technology precisely aligns the position and orientation of a single-direction LiDAR and a camera, enabling accurate sensor alignment even in environments with strict weight and space constraints. As a result, data fusion between sensors becomes possible, allowing robots or drones to perform various missions such as autonomous flight, exploration, and disaster rescue more reliably.

Ultracompact drones are highly useful for disaster site rescue, industrial facility inspection, and exploration of confined spaces. However, due to strict weight limits on sensors that can be mounted on the drone, ultralight sensors such as single-direction LiDAR must be used instead of high-performance 3D LiDAR sensors.

In such cases, the amount of data obtained is extremely limited, making it much more difficult to precisely match the coordinate systems between sensors-similar to assembling a puzzle with only a few pieces. This increases the importance of calibration technology.

3D LiDAR is a sensor technology that emits lasers in multiple directions, measures the reflection times, and generates three-dimensional point cloud data of the surrounding environment. It is used for precise spatial recognition in applications such as autonomous driving, robotics, and drones.

Conventional LiDAR-camera calibration methods have relied on dense point cloud data obtained from multi-beam 3D LiDAR. However, in ultracompact drones, only single-direction LiDAR can be used, resulting in extremely sparse data. In other words, since the puzzle pieces needed to complete the picture are scattered, it is difficult to calibrate using existing methods.

To solve this problem, the research team utilized the Manhattan World model, a technique that simplifies calculations by assuming that most buildings and indoor structures are aligned horizontally and vertically. The team also proposed a new approach using a custom-designed calibration board, which consists of a black-and-white grid and small square holes and can rotate like a door.

Although LiDAR originally measures distance values in only one direction, by utilizing the changes in distance as the board rotates, along with the information from the holes and the board’s shape, a single measurement point can be expanded into three reference points.

To validate their method, the research team mounted sensors on the nanodrone platform 'Crazyflie' (9cm x 9cm, 27g) and conducted various experiments. As a result, the proposed method reduced the mean reprojection error-a key indicator of sensor calibration accuracy-to about 3 pixels, demonstrating significantly higher precision compared to conventional methods. In addition, the position error between sensors was about 1cm, and the angular error for three-axis rotation (roll, pitch, and yaw) was only about 1°, proving stable calibration performance even in ultracompact drones.

'SPLiCE' achieved high accuracy with just 15 data pairs, far fewer than the checkerboard-based or virtual 2D LiDAR-based methods, which typically require more than 40 pairs. This greatly improves efficiency and practicality compared to existing methods.

With these results, stable sensor calibration is now possible even in ultracompact drones with strict weight and power constraints, and this technology could serve as a key foundation for various future autonomous flight applications.

Professor Kim Pyojin stated, "This research provides a technology that enables precise calibration of LiDAR and cameras even in environments with strict sensor weight and performance limitations, such as ultracompact drones. In the future, it is expected to be used for missions such as mapping and exploration performed by multiple drones working together."

This content was produced with the assistance of AI translation services.