Granted Patent B2

US 12,573,060 · App. 18/340,318 · Granted Mar 10, 2026

Inventors: Ken Kim (Seoul, KR); Ji Wuck Jung (Goyang-si, KR)

Assignee: 3I INC.

G06T7/248G06V10/70H04N23/61H04N23/611H04N23/632H04N23/633G06T2200/24G06T2207/10016G06T2207/20081G06T2207/30244

Abstract

Claims (58)

1 . A locked-on target based object tracking method performed in a portable terminal and provides an object tracking function to a recorded video captured for a forward view of the portable terminal, the method comprising:

setting a tracking object;

identifying an object which is existing in a frame image of the recorded video using a first artificial intelligence model and setting the object as a selected object, wherein the object and the tracking object belong to the same type; and

determining whether the selected object set by the first artificial intelligence model and the tracking object are the same object using a second artificial intelligence model trained in a different way from the first artificial intelligence model,

wherein the second artificial intelligence model is an artificial intelligence model pairwise-trained with a default image and a training image modified based on the default image to generate feature data representing identity between a reference object in the default image and a comparison object in the training image, and

wherein the determining comprises:

generating a first feature vector for a reference object corresponding to the tracking object in the default image and a second feature vector for a comparison object corresponding to the selected object in the training image; and

determining that the selected object and the tracking object are the same object when a similarity between the vectors meets a preset threshold.

2 . The locked-on target based object tracking method of claim 1 , wherein the first artificial intelligence model is an artificial intelligence model that is deep-learned with a plurality of training image sets corresponding to the type of the tracking object and identifies an object corresponding to the type of the tracking object existing in the frame image.

3 . The locked-on target based object tracking method of claim 1 , wherein the setting of the selected object includes:

displaying a bounding box for the selected object identified by the first artificial intelligence model; and

cropping the bounding box and inputting a cropped bounding box to the second artificial intelligence model.

4 . The locked-on target based object tracking method of claim 3 , wherein the determining of whether the tracking object is the same object includes:

generating first feature data for a first bounding box including the selected object using the second artificial intelligence model; and

generating second feature data for a second bounding box including the tracking object using the second artificial intelligence model.

5 . The locked-on target based object tracking method of claim 4 , wherein the determining of whether the tracking object is the same object further includes comparing the first feature data and the second feature data to determine the selected object and the tracking object belong to the same type.

6 . The locked-on target based object tracking method of claim 1 , wherein the default image and the training image are images including a person's whole body appearance.

7 . The locked-on target based object tracking method of claim 1 , further comprising:

when the selected object is the same object as the tracking object, resetting the selected object as the tracking object and performing object tracking based on the reset tracking object.

8 . The locked-on target based object tracking method of claim 7 , wherein the performing of the object tracking based on the reset tracking object includes:

setting a resolution of a viewing window to have a second resolution lower than a first resolution of the frame image; and

setting a partial area of the selected frame image including the reset tracking object as the viewing window based on the reset tracking object.

9 . The locked-on target based object tracking method of claim 8 , wherein the performing of the object tracking based on the reset tracking object further includes displaying the viewing window based on a user display interface.

10 . A portable terminal comprising:

a memory configured to store one or more instructions; and

at least one processor configured to execute the one or more instructions stored in the memory,

wherein the at least one processor executes the one or more instructions to:

receive a recorded video captured at a first resolution from a camera unit;

set a tracking object;

identify an object which is existing in a frame image of the recorded video using a first artificial intelligence model and set the object as a selected object, wherein the object and the tracking object belong to the same type; and

determine whether the selected object set by the first artificial intelligence model and the tracking object are the same object using a second artificial intelligence model trained in a different way from the first artificial intelligence model,

wherein the second artificial intelligence model is an artificial intelligence model pairwise-trained with a default image and a training image modified based on the default image to generate feature data representing identity between a reference object in the default image and a comparison object in the training image, and

wherein the determining comprises:

generating a first feature vector for a reference object corresponding to the tracking object in the default image and a second feature vector for a comparison object corresponding to the selected object in the training image; and

determining that the selected object and the tracking object are the same object when a similarity between the vectors meets a preset threshold.

11 . The portable terminal of claim 10 , wherein the first artificial intelligence model is an artificial intelligence model that is deep-learned with a plurality of training image sets corresponding to a type of the tracking object and identifies an object corresponding to the type of the tracking object existing in the frame image.

12 . The portable terminal of claim 10 , wherein, when setting the selected object, the portable terminal displays a bounding box for the selected object identified by the first artificial intelligence model, and crops the bounding box and inputs a cropped bounding box to the second artificial intelligence model.

13 . The portable terminal of claim 12 , wherein, when determining whether the tracking object is the same object, the portable terminal generates:

first feature data for a first bounding box including the selected object using the second artificial intelligence model; and

second feature data for a second bounding box including the tracking object using the second artificial intelligence model.

14 . The portable terminal of claim 13 , wherein, when determining whether the tracking object is the same object, the portable terminal compares the first feature data and the second feature data to determine the selected object and the tracking object belong to the same type.

15 . The portable terminal of claim 10 , wherein the default image and the training image are images including a person's whole body appearance.

16 . The portable terminal of claim 10 , wherein when the selected object is the same object as the tracking object, the portable terminal resets the selected object as the tracking object and performs object tracking based on the reset tracking object.

17 . The portable terminal of claim 16 , wherein, when performing the object tracking based on the reset tracking object, the portable terminal sets:

a resolution of a viewing window to have a second resolution lower than a first resolution of the frame image; and

a partial area of the selected frame image including the reset tracking object as the viewing window based on the reset tracking object.

18 . A non-transitory storage medium in which computer-readable instructions are stored, wherein, when executed by a portable terminal, the instructions cause the portable terminal to:

set a tracking object,

identify an object which is existing in a frame image of a recorded video using a first artificial intelligence model and set the object as selected object, wherein the object and the tracking object belong to the same type, and

determine whether the selected object set by the first artificial intelligence model and the tracking object are the same object using a second artificial intelligence model trained in a different way from the first artificial intelligence model,

wherein the second artificial intelligence model is an artificial intelligence model pairwise-trained with a default image and a training image modified based on the default image to generate feature data representing identity between a reference object in the default image and a comparison object in the training image, and

wherein the determining comprises:

generating a first feature vector for a reference object corresponding to the tracking object in the default image and a second feature vector for a comparison object corresponding to the selected object in the training image; and

determining that the selected object and the tracking object are the same object when a similarity between the vectors meets a preset threshold.

19 . The locked-on target based object tracking method of claim 1 , wherein the second artificial intelligence further generates a first feature vector for the reference object and a second feature vector for the comparison object, respectively, and determines whether the comparison object is similar to the reference object using the first feature vector of the reference object and the second feature vector of the comparison object.

20 . The portable terminal of claim 10 , wherein the second artificial intelligence is further configured to:

generate a first feature vector for the reference object and a second feature vector for the comparison object, respectively; and

determine whether the comparison object is similar to the reference object using the first feature vector of the reference object and the second feature vector of the comparison object.

Assignments (1)

Priority Claims (4)

Continuity (3)

References Cited (35)