US 12,430,867 B2
DNN-based object recognition method for multi-channel fisheye images of AVM video
Sang Gu Kim, Seoul (KR); Jin Bok Kim, Seongnam-si (KR); and In Sub Lee, Seongnam-si (KR)
Assigned to LITBIG INC., Seongnam-si (KR)
Filed by Sang Gu Kim, Seoul (KR); Jin Bok Kim, Seongnam-si (KR); and In Sub Lee, Seongnam-si (KR)
Filed on Jan. 22, 2025, as Appl. No. 19/034,330.
Claims priority of application No. 10-2024-0026234 (KR), filed on Feb. 23, 2024.
Prior Publication US 2025/0272941 A1, Aug. 28, 2025
Int. Cl. H04N 7/00 (2011.01); G06V 10/10 (2022.01); G06V 10/82 (2022.01); G06V 20/40 (2022.01); G06V 20/58 (2022.01)
CPC G06V 10/16 (2022.01) [G06V 10/82 (2022.01); G06V 20/40 (2022.01); G06V 20/58 (2022.01)] 5 Claims
OG exemplary drawing
 
1. A DNN-based object recognition method for multi-channel fisheye images of AVM video in a vehicle AVM apparatus, comprising:
obtaining multi-channel unit images which are produced by fisheye-lens cameras of the vehicle AVM apparatus;
forming a cylindrical projection plane around each of the fisheye-lens cameras;
projecting the multi-channel unit images onto the cylindrical projection planes respectively so as to obtain a plurality of unit projection images;
combining the unit projection images on a single image so as to form a composite projection image;
inputting the composite projection image to a pre-trained Deep Neural Network (DNN) model so as to obtain a composite object-recognition output;
decomposing the composite object-recognition output by the layout of the unit projection images so as to obtain unit object-recognition output for each of the multi-channel unit images;
identifying an object for each direction of the vehicle out of the unit object-recognition output; and
calculating a gap distance to the identified object based on relative position in a horizontal plane between the bottom part of the identified object and the corresponding fisheye-lens camera with assuming that the bottom part of the identified object is attached to the ground.