US 12,260,666 B2
Apparatus and method for identifying condition of animal object based on image
Kwang Myung Jeon, Gwangju (KR); and In Chul Ryu, Gwangju (KR)
Assigned to INTFLOW INC., Gwangju (KR)
Filed by INTFLOW INC., Gwangju (KR)
Filed on Mar. 23, 2022, as Appl. No. 17/702,417.
Claims priority of application No. 10-2021-0105322 (KR), filed on Aug. 10, 2021.
Prior Publication US 2023/0049090 A1, Feb. 16, 2023
Int. Cl. G06V 40/10 (2022.01); A01K 29/00 (2006.01); G06T 7/00 (2017.01); G06T 7/73 (2017.01); G06V 10/44 (2022.01); G06V 10/46 (2022.01); G06V 10/764 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06V 40/10 (2022.01) [A01K 29/00 (2013.01); G06T 7/0012 (2013.01); G06T 7/73 (2017.01); G06V 10/457 (2022.01); G06V 10/46 (2022.01); G06V 10/764 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06T 2207/30188 (2013.01); G06T 2207/30232 (2013.01)] 15 Claims
OG exemplary drawing
 
1. An image-based animal object condition identification apparatus, comprising:
a communication module that receives an image of an object;
a memory that stores therein a program configured to extract animal condition information from the received image; and
a processor that executes the program,
wherein the program extracts continuous animal detection information of each object by inputting the received image into an animal detection model that is trained based on learning data composed of animal images and outputs predetermined animal condition information for each class of each animal object by inputting the continuous animal detection information of each object into an animal condition identification model, and
the animal detection information is extracted from n number of continuous entire images including at least one animal object, and includes n number of continuous object images and n number of continuous object detection data corresponding to the respective object images,
wherein the animal condition identification model is constructed based on the n number of continuous entire images including at least one animal object and learning data in which the animal condition information is matched with each class of each animal object included in each of the continuous entire images, and
the animal condition identification model includes:
a first feature extraction unit that generates n number of one-dimensional image data by converting the n number of continuous object images into monochrome images and generates feature data of a first length based on the one-dimensional image data by using a convolutional neural network (CNN);
a second feature extraction unit that generates n number of one-dimensional data of a second length by connecting the n number of continuous object detection data and generates feature data of the second length based on the one-dimensional image data of the second length by using a first feed-forward neural network (FFNN); and
an output unit that generates data of a third length by connecting the feature data of the first length and the feature data of the second length and outputs the animal condition information based on the data of the third length by using a second FFNN.