Granted Patent B2

US 12,541,952 · App. 18/132,183 · Granted Feb 3, 2026

Inventors: Jin Sol Kim (Hwaseong-si, KR); Se Jeong Lee (Suwon-si, KR); Jung Woo Heo (Suwon-si, KR)

Assignees: HYUNDAI MOTOR COMPANY; KIA CORPORATION

G06V10/774G06V10/764G06V10/778G06V10/95G06V20/56

Abstract

Claims (55)

1 . An apparatus for collecting a training image of a deep learning model, the apparatus comprising:

a camera sensor configured to capture an image of a surrounding of a vehicle; and

a controller configured to:

determine an entropy of each pixel in the image,

determine an entropy of each class in the image based on the entropy of each pixel in the image, and

determine whether to collect the image based on the entropy of each class,

wherein the controller is configured to:

determine pixels constituting a first class in the image; and

determine, as entropy of the first class, an entropy average of the pixels constituting the first class.

2 . The apparatus of claim 1 , wherein the controller is configured to:

collect the image as the training image of the deep learning model when the entropy of each class is within a preset range.

3 . The apparatus of claim 2 , wherein the controller is configured to:

determine a priority of the training images collected, and

store a preset number of training images in a storage based on the priority.

4 . The apparatus of claim 3 , wherein the controller is configured to:

determine the entropy average of each class included in each training image, and

assign a highest priority to a training image having a highest entropy average.

5 . The apparatus of claim 3 , wherein the controller is configured to:

determine an entropy of a specific class among classes included in each training image, and

assign a highest priority to a training image having a highest entropy of the specific class.

6 . The apparatus of claim 3 , wherein the controller is configured to:

determine, when the training images are additionally collected in a state that the preset number of training images are stored in the storage, the preset number of training images in a higher priority order with respect to the training images stored and the training images additionally collected.

7 . The apparatus of claim 6 , wherein the controller is configured to:

substitute the preset number of training images stored in the storage with the preset number of training images determined.

8 . The apparatus of claim 3 , wherein the controller is configured to:

periodically transmit, to a wireless communication device in a start-off state of the vehicle, the training images stored in the storage.

9 . The apparatus of claim 8 , wherein the controller is configured to:

transmit, to the wireless communication device in a start-on state of the vehicle, a message for requesting to prepare for receiving the training images, and

receive, from the wireless communication device in the start-off state of the vehicle, a message for indicating that preparing for receiving the training images is completed.

10 . A method for collecting a training image of a deep learning model, the method comprising:

taking, by a camera sensor, an image of a surrounding of a vehicle;

determining, by a controller, an entropy of each pixel in the image;

determining, by the controller, an entropy of each class in the image based on the entropy of each pixel in the image; and

determining, by the controller, whether to collect the image based on the entropy of each class,

wherein determining the entropy of the class in the image includes:

determining, by the controller, pixels constituting a first class in the image, and

determining, by the controller, as entropy of the first class, an entropy average of pixels constituting the first class.

11 . The method of claim 10 , wherein determining whether to collect the image includes:

collecting, by the controller, the image as the training image of the deep learning model when all entropy of each class satisfies a preset range.

12 . The method of claim 11 , wherein determining whether to collect the image further includes:

determining, by the controller, a priority of the training images collected, and

storing, by the controller, a preset number of training images in a storage based on the priority.

13 . The method of claim 12 , wherein determining the priority of the training images collected include:

determining, by the controller, the entropy average of each class included in each training image, and

assigning, by the controller, a highest priority to a training image having a highest entropy average.

14 . The method of claim 12 , wherein determining the priority of the training images collected includes:

determining, by the controller, an entropy of a specific class among classes included in each training image, and

assigning, by the controller, a highest priority to a training image having a highest entropy of the specific class.

15 . The method of claim 12 , wherein storing the preset number of the training images further includes:

determining, by the controller, when the training images are additionally collected in a state that the preset number of the training images are stored in the storage, the preset number of training images in a higher priority order with respect to the training images stored and the training images additionally collected, and

substituting, by the controller, the preset number of training images stored in the storage with the preset number of training images determined.

16 . The method of claim 12 , wherein determining whether to collect the image further includes:

transmitting, by the controller to a wireless communication device in a start-on state of the vehicle, a message for requesting to prepare for receiving the training images;

receiving, by the controller from the wireless communication device, in a start-off state of the vehicle, a message for indicating that preparing for receiving the training images is completed; and

periodically transmitting, by the controller to the wireless communication device in the start-off state of the vehicle, the training images stored in the storage.

Assignments (1)

Priority Claims (1)

Continuity (1)

References Cited (14)