# US 12,112,262 B2 · Real-time prediction program for change in occupants in large exhibition hall based on deep learning

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

- Number: US 12,112,262 B2
- Document ID: US12112262B2
- Status: Granted
- Kind: B2
- Application number: 17/092,782
- Filed: 2020-11-09
- Published: 2021-06-10
- Granted: 2024-10-08
- Earliest claimed: 2019-12-10
- Assignees: Pusan National University Industry-University Cooperation Foundation
- Applicants: Pusan National University Industry-University Cooperation Foundation
- Inventors: Giltae Song; Seonghyeon Kim
- Web page: https://patentgrep.com/patent/US12112262B2

## CPC

- G06N 3/08: Learning methods
- G06F 18/10: Pre-processing; Data cleansing
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/09: Supervised learning
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/776: Validation; Performance evaluation
- G06V 10/82: using neural networks
- G06V 20/36: {Indoor scenes}
- G06V 20/52: Surveillance or monitoring of activities, e.g. for recognising suspicious objects (recognising microscopic objects G06V20/69)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Disclosed are a method and apparatus for predicting a change in the occupants within a large exhibition hall in real time based on deep learning. A proposed method of predicting a change in the number of occupants within a space in real time includes dividing, into zones, a space where a number of occupants is to be predicted and pre-processing data related to a number of occupants within the space collected through simulations, generating the pre-processed data in a form of time-series data for deep learning, training a deep learning model for predicting a number of occupants in each divided zone using the generated time-series data, and predicting the number of occupants within the space by inputting, to the trained model, the data related to a number of occupants within the space collected in real time through socket communication with a server.

## Claims

### Claim 1

A method of predicting a change in the number of occupants within a space in real time using a computer device, the method comprising:
dividing, by at least one processor of the computer device, into zones, a space where a number of occupants is to be predicted and pre-processing collected data related to the number of occupants within the space through simulations;
generating, by the at least one processor, the pre-processed data in a form of time-series data for deep learning;
training, by the at least one processor, a deep learning model for predicting a number of occupants in each of the divided zones using the generated time-series data; and
predicting, by the at least one processor, the number of occupants within the space by inputting, to the trained model, real time data related to the number of occupants within the space collected in real time through socket communication with a server;
wherein the training of the deep learning model comprises:
using an individual model configuring a share prediction model for receiving the time-series data based on an LSTM and outputting one occupancy number in one zone on a specific time; and
training the individual model using the pre-processed data with regard to each of the divided zones, respectively,
wherein a prediction of a share for the space comprising the divided zones is determined as a weighted sum of each share for each of the divided zones on a current time.

### Claim 2 (depends on 1)

The method of claim 1, wherein the dividing comprises:
measuring a share of visitors to each zone using image sensors and counting devices installed at given intervals, and
reinforcing the collected data through simulations in order to train the deep learning model.

### Claim 3 (depends on 2)

The method of claim 2, wherein:
the share of the visitors to each divided zone is measured for a predetermined time interval and a predetermined time, and
all of the time-series data sets are normalized within a range of 0 to 1 using a zero-padding scheme and minimum-maximum normalization.

### Claim 4 (depends on 1)

The method of claim 1, wherein the generating the pre-processed data in a form of time-series data for deep learning comprises re-configuring the pre-processed data based on the deep learning model using a long-short term memory (LSTM).

### Claim 5

A computer apparatus for predicting a change in a number of occupants within a space in real time, the apparatus comprising:
a pre-processor configured to pre-process collected data related to a number of occupants within a space through simulations;
a time-series data generator configured to generate the pre-processed data in a form of time-series data for deep learning;
a training unit configured to divide, into zones, a space where a number of occupants is to be predicted and to train a deep learning model for predicting a number of occupants in each of the divided zones using the generated time-series data; and
a prediction unit configured to predict the number of occupants within the space by inputting, to the trained model, real time data related to the number of occupants within the space collected in real time through socket communication with a server;
wherein the training unit is configured to:
use an individual model configuring a share prediction model for receiving the time-series data based on an LSTM and outputting one occupancy number in one zone on a specific time; and
train the individual model using the pre-processed data with regard to each of the divided zones, respectively;
wherein a prediction of a share for the space comprising the divided zones is determined as a weighted sum of each share for each of the divided zones on a current time.

### Claim 6 (depends on 5)

The apparatus of claim 5, wherein the pre-processor is configured to:
measure a share of visitors to each zone using image sensors and counting devices installed at given intervals, and
reinforce the collected data through simulations in order to train the deep learning model.

### Claim 7 (depends on 6)

The apparatus of claim 6, wherein:
the share of the visitors to each divided zone is measured for a predetermined time interval and a predetermined time, and
all of the time-series data sets are normalized within a range of 0 to 1 using a zero-padding scheme and minimum-maximum normalization.

### Claim 8 (depends on 5)

The apparatus of claim 5, wherein the time-series data generator is configured to re-configure the pre-processed data based on the deep learning model using a long-short term memory (LSTM).