# US 2020/0005086 A1 · Deep learning-based automatic gesture recognition method and system

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0005086 A1
- Document ID: US20200005086A1
- Status: Published application
- Kind: A1
- Application number: 16/147,962
- Filed: 2018-10-01
- Published: 2020-01-02
- Earliest claimed: 2018-06-29
- Assignees: Korea Electronics Technology Institute
- Applicants: Korea Electronics Technology Institute
- Inventors: Sang Ki KO; Choong Sang CHO; Hye Dong JUNG; Young Han LEE
- Web page: https://patentgrep.com/patent/US20200005086A1

## CPC

- G06V 10/32: Normalisation of the pattern dimensions
- G06F 18/2148: {characterised by the process organisation or structure, e.g. boosting cascade}
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06V 10/44: Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/7747: {Organisation of the process, e.g. bagging or boosting}
- G06V 10/776: Validation; Performance evaluation
- G06V 10/82: using neural networks
- G06V 40/20: Movements or behaviour, e.g. gesture recognition (recognition of facial expressions G06V40/16)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Deep learning-based automatic gesture recognition method and system are provided. The training method according to an embodiment includes: extracting a plurality of contours from an input image; generating training data by normalizing pieces of contour information forming each of the contours; and training an AI model for gesture recognition by using the generated training data. Accordingly, robust and high-performance automatic gesture recognition can be performed, without being influenced by an environment and a condition even while using less training data.

## Claims

### Claim 1

A training method, comprising:
extracting a plurality of contours from an input image;
generating training data by normalizing pieces of contour information forming each of the contours; and
training an AI model for gesture recognition by using the generated training data.

### Claim 2 (depends on 1)

The training method of claim 1, wherein the contours overlap one another.

### Claim 3 (depends on 1)

The training method of claim 1, wherein the pieces of contour information are pieces of information regarding feature points extracted to derive the contours.

### Claim 4 (depends on 3)

The training method of claim 3, wherein the pieces of information regarding the feature points comprise pieces of coordinate information of the feature points.

### Claim 5 (depends on 4)

The training method of claim 4, wherein the generating the training data comprises normalizing the pieces of contour information through arithmetic operations using a mean of the pieces of coordinate information of the pieces of contour information forming each contour, and a standard deviation.

### Claim 6 (depends on 1)

The training method of claim 1, wherein the generating the training data comprises generating the training data by adding pieces of reliability information of the pieces of contour information to the normalized pieces of coordinate information.

### Claim 7 (depends on 1)

The training method of claim 1, further comprising:
extracting feature data from each of regions comprising the contours; and
adding the extracted feature data to the generated training data,
wherein the training comprises training the AI model by using the training data to which the feature data is added.

### Claim 8 (depends on 7)

The training method of claim 7, wherein the extracting comprises extracting the feature data by using a neural network.

### Claim 9 (depends on 7)

The training method of claim 7, wherein a dimension of the feature data is identical to a dimension of the pieces of contour information.

### Claim 10 (depends on 1)

The training method of claim 1, further comprising augmenting the generated training data,
wherein the training comprises training the AI model by using the augmented training data.

### Claim 11 (depends on 10)

The training method of claim 10, wherein the augmenting comprises additionally generating training data by randomly changing the normalized contour information forming the training data.

### Claim 12 (depends on 11)

The training method of claim 11, wherein the augmenting comprises changing the normalized contour information within a limited range.

### Claim 13 (depends on 12)

The training method of claim 12, wherein the limited range is a contour region comprising the contour information.

### Claim 14 (depends on 13)

The training method of claim 13, wherein the limited range is a limited distance from the contour information to another contour information.

### Claim 15

A gesture recognition system comprising:
an input unit configured to receive an image; and
a processor configured to extract a plurality of contours from an image inputted through the input unit, to generate training data by normalizing pieces of contour information forming each of the contours, and to train an AI model for gesture recognition by using the generated training data.

### Claim 16

A gesture cognition method comprising:
extracting a plurality of contours from an input image;
generating inference data by normalizing pieces of contour information forming each of the contours; and
recognizing a gesture by inputting the generated inference data to an AI model for gesture recognition.
