The novel method, designed by an engineer at Panasonic, estimates and corrects fisheye distortion with a device like a smartphone or a camera. First reported on by Nikkei, Pansonic claims to have solved fisheye distortion.
The issue of lens distortion
Generally, processing and correcting distortion requires multiple images and a computer. The tilt of the lens informs the level of distortion in images. Soft and hardware tend to be too slow or inefficient to solve it, especially with a single image.
The new method, created by Panasonic engineer Nobuhiko Wakai, uses “learning-based calibration” to address tilt angle, roll angle, focal length, and distortion in-camera. More details on the topic can be found in Panasonic and Wakai’s abstract.
A new method two years in the making
Wakai has been developing the method at Panasonic with a team for two years. Fisheye images, the abstract explains, are degraded by “mismatching between the actual projection and expected projection.”
Recent similar methods, Wakai allows, tend to use a single image to predict extrinsic and intrinsic camera parameters. Still, fisheye images degrade the quality. Wakai and his team have been seeking to solve this issue specifically.
A generic camera model and learning-based methods
Wakai and his team “propose a generic camera model that has the potential to address various types of distortion.” Further, they suggest their learning-based calibration method be used with this model. Extensive testing and datasets showed that their method outperformed other conventional methods.
The future of the method
“To improve the calibration performance in off-the-shelf cameras,” concludes the abstract, “in future work, we will study the dataset domain mismatch.”
Panasonic intends to do further practical tests. Should the method remain successful, the company intends to move forward with its sale.