Kathan ShuklaFollow


A vehicle system may use a computer vision module to rectify distortions in images captured by one or more fish-eye cameras (hereinafter referred to as “fish-eye camera images”) of the surroundings of a vehicle (e.g., an automobile, a motorcycle, a bus, a recreational vehicle (RV), a semi-trailer truck, a tractor or other type of farm equipment, train, a plane, a boat, a helicopter, a personal transport vehicle, etc.). In some examples, a machine-learning model (hereinafter referred to as “ML model”) may calibrate (or, in other words, tune) intrinsic parameters (e.g., the fish-eye camera images’ focal length ((x,y) coordinates), principal point ((x,y) coordinates), field of view (FOV) scale, etc.) used by the computer vision module. The ML model may calibrate the intrinsic parameters by identifying, in fish-eye camera images, specific points (e.g., elements, features, objects, etc.) with well-defined and known geometries in reality (e.g., such as straight lines, square corners, etc.) that have been distorted by the fish-eye cameras. The ML model may be trained to determine intrinsic parameters that correspond to a geometrical transformation that, when applied to the fish-eye camera images, dewarps the specific points such that the dewarped images accurately represent the specific points’ geometries. The computer vision module may then generate maps that denote this geometrical transformation and apply the geometrical transformation to the fish-eye camera images to produce dewarped images.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.