US 12,354,298 B2
Image processing apparatus and vehicle
Yusuke Goto, Tokyo (JP); and Minoru Kuraoka, Tokyo (JP)
Assigned to SUBARU CORPORATION, Tokyo (JP)
Filed by SUBARU CORPORATION, Tokyo (JP)
Filed on Sep. 12, 2022, as Appl. No. 17/942,241.
Claims priority of application No. 2021-155975 (JP), filed on Sep. 24, 2021.
Prior Publication US 2023/0098659 A1, Mar. 30, 2023
Int. Cl. G06T 7/70 (2017.01); G05D 1/00 (2006.01); G05D 1/02 (2020.01); G06T 3/60 (2024.01); G06V 10/44 (2022.01); G06V 20/56 (2022.01)
CPC G06T 7/70 (2017.01) [G05D 1/021 (2013.01); G06T 3/60 (2013.01); G06V 10/443 (2022.01); G06V 20/56 (2022.01); G06T 2207/20081 (2013.01)] 5 Claims
OG exemplary drawing
 
1. An image processing apparatus mountable on a vehicle and comprising:
one or more processors; and
one or more memories communicably coupled to the one or more processors, in which the one or more processors are configured to:
extract a feature quantity included in a captured image; and
perform identification of an object on a basis of the feature quantity, wherein
the one or more processors are configured to extract the feature quantity by performing, on a basis of the captured image, a convolution calculation multiple times using a filter including multiple filter values that are two-dimensionally arranged,
the multiple filter values of the filter are initially set before starting machine learning at values that are horizontally symmetric with respect to a vertical axis of symmetry along a horizontal direction, and
augmentation batch data of learning data configured to be used in an update process of the filter by the machine learning consists of one or more unflipped images and one or more flipped images as paired images, each flipped image resulting from applying image-flipping to a corresponding one of the unflipped images with respect to the vertical axis of symmetry such that the each flipped image is a horizontally flipped image from the corresponding one of the unflipped images, wherein
the one or more processors are configured to secure horizontal symmetry regarding:
a result of the identification of the object in a case where a travel environment for the vehicle is a left-side travel environment, and
a result of the identification of the object in a case where the travel environment for the vehicle is a right-side travel environment, and
the one or more processors are configured to set the multiple filter values of the filter that are initially set before starting the machine learning by updating every two filter values of the multiple filter values located at two horizontally symmetric positions with respect to the vertical axis of symmetry and having different values to an average value of the two filter values.