KR-20260062255-A - DETECTING Method OF LOCATION BASED ON ARTIFICIAL INTELLEGENCE MODEL and A SERVER DEVICE supporting the same

KR20260062255AKR 20260062255 AKR20260062255 AKR 20260062255AKR-20260062255-A

Abstract

An embodiment of the present invention discloses an AI model-based position detection method and a server device supporting the same, wherein a server processor of a server device is configured to perform the steps of: calculating a PDR trajectory according to the movement of a portable electronic device based on PDR sensor information and a position information trajectory according to the movement of a portable electronic device based on GPS sensor information; aligning the PDR trajectory to the position information trajectory; labeling the PDR sensor information included in the PDR trajectory aligned to the position information trajectory; and performing attitude angle correction of the portable electronic device for the labeled information to generate first training data.

Inventors

이태경

Assignees

에스케이텔레콤 주식회사

Dates

Publication Date: 20260507
Application Date: 20241028

Claims (10)

Server memory storing GPS (global positioning system) sensor information and PDR (Pedestrian Dead Reckoning) sensor information collected according to the movement of a portable electronic device; A server processor functionally connected to the above server memory; including, The above server processor is, Calculate the PDR trajectory according to the movement of the portable electronic device based on the above PDR sensor information and the location information trajectory according to the movement of the portable electronic device based on the above GPS sensor information, and Align the PDR trajectory with the above position information trajectory, and Labeling is performed on the PDR sensor information included in the PDR trajectory aligned with the above position information trajectory, and A server device supporting position detection based on an AI (artificial intelligence) model, characterized by being configured to generate first training data by performing attitude angle correction of the portable electronic device for the above-mentioned labeled information.
In paragraph 1, The above server processor is, A server device supporting AI model-based location detection, characterized by being configured to generate an updated first PDR AI model by retraining a previously trained PDR AI model based on the first training data.
In paragraph 2, The above server processor is, A server device supporting AI model-based location detection, characterized by being configured to generate second training data by repeating the process of generating the first training data, generate an updated second PDR AI model by retraining the prior-trained PDR AI model based on the second training data, and then perform error comparison by inputting pre-set test data into the updated first PDR AI model and the updated second PDR AI model.
In paragraph 3, The above server processor is, If, as a result of the error comparison above, the result of the second PDR AI model is improved compared to the result of the first PDR AI model, the process of generating the first training data is repeated to generate third training data and generate an updated third PDR AI model, and A server device supporting AI model-based location detection, characterized in that, as a result of the error comparison above, if the result of the second PDR AI model shows an error equal to or greater than the result of the first PDR AI model, the updated first PDR AI model is configured to be distributed to the portable electronic device.
A server processor of a server device calculates a PDR (Pedestrian Dead Reckoning) trajectory based on PDR sensor information and a position information trajectory based on GPS (global positioning system) sensor information based on the movement of the portable electronic device; A step of aligning the PDR trajectory with the above position information trajectory; A step of performing labeling on PDR sensor information included in the PDR trajectory aligned with the above position information trajectory; An AI model-based position detection method characterized by being configured to perform the step of generating first training data by performing attitude angle correction of the portable electronic device for the above-mentioned labeled information.
In paragraph 5, A method for detecting a location based on an AI model, further comprising the step of retraining a prior-trained PDR AI model based on the first training data to generate an updated first PDR AI (artificial intelligence) model.
In paragraph 6, A step of generating second training data by repeatedly performing the process of generating the first training data; A step of generating an updated second PDR AI model by retraining the prior-trained PDR AI model based on the second training data; An AI model-based position detection method characterized by further including the step of performing an error comparison by inputting pre-set test data into the updated first PDR AI model and the updated second PDR AI model.
In Paragraph 7, A method for detecting a location based on an AI model, further comprising the step of generating a third PDR AI model by repeating the process of generating the first training data to generate the third training data and generating an updated third PDR AI model when the result of the second PDR AI model is improved compared to the result of the first PDR AI model as a result of the error comparison above.
In Paragraph 7, A method for detecting a position based on an AI model, further comprising the step of distributing the updated first PDR AI model to the portable electronic device when, as a result of the error comparison above, the result of the second PDR AI model shows an error equal to or greater than the result of the first PDR AI model.
As a computer program stored on a computer-readable recording medium, The above computer program is, A computer program comprising instructions for a processor to perform a method according to any one of paragraphs 5 through 9.

Description

AI Model-Based Location Detection Method and Server Device Supporting the Same The present invention relates to performing location detection of a portable electronic device using an AI model. Portable electronic devices provide location detection functions for various purposes. For example, portable electronic devices provide location detection functions to locate children or the elderly. Additionally, portable electronic devices provide location detection functions related to navigation features that support movement to a user-designated target point. In this regard, conventional portable electronic devices have been equipped with a Global Positioning System (GPS) sensor and have provided the function of detecting location by receiving GPS signals from multiple GPS satellites. However, portable electronic devices may be placed in situations where they cannot receive at least some of the GPS signals due to their current surrounding environment. For instance, when entering areas such as valleys or buildings where GPS signal reception is difficult, the portable electronic device may be unable to receive multiple GPS signals or may receive fewer than a certain number, making accurate location detection difficult. To enable location detection in such environments where GPS signal reception is difficult, conventional devices have supported Pedestrian Dead Reckoning (PDR) sensing-based location detection functions—that is, PDR technology—which measure the movement of the user carrying the portable electronic device using the device's sensors and detect the location based on this data. For example, given a user's initial location and orientation indoors, the current user's location can be estimated by calculating and accumulating position changes using only the internal sensor information (or sensor data) of the portable electronic device, based on technologies such as network fingerprinting. These PDR technologies include sensor integration and motion pattern analysis methods. The aforementioned sensor integration method estimates the user's speed, distance, and direction by integrating data obtained from accelerometers and gyroscopes over time. However, in this method, errors in sensor information accumulate over time, which can cause speed and position to diverge instantaneously. Consequently, a zero-update process is essential to periodically detect and correct for stationary states. In the case of sensor integration methods utilizing sensors mounted on shoes, a stationary state with the shoe in contact with the ground exists with every step, allowing the zero-update process to be performed. However, for portable electronic devices, a stationary state does not exist during walking, making it nearly impossible to apply zero-update correction. The aforementioned motion pattern analysis method updates the position by estimating the number of steps, stride length, and direction through sensor information analysis and accumulating them. This method has the advantage of preventing the problem of rapid divergence in the estimated position. However, since the motion pattern analysis method is designed to operate only for specific portable electronic device holding postures (handheld, phone-call, pocket, swing, backpack, etc.), it may fail to estimate the position if the posture changes. As a solution to this, one could consider using a separate classifier that recognizes the posture of holding the portable electronic device and executing a separate algorithm tailored to that specific posture; however, it is practically difficult to distinguish between recognizing intervals where the posture changes and recognizing various general postures. To overcome the limitations of the aforementioned PDR technology, research on AI model-based PDR technology is being conducted. The input to the PDR AI model is sensor information collected by sensors (accelerometers, gyroscopes, magnetometers, etc.) over a short period (e.g., 1 second), and the output of the PDR AI model is the pedestrian's velocity (direction of movement, speed). By continuously integrating the velocity output of the PDR AI model, the pedestrian's position can be estimated. If sensor information from various postures and situations involving changes in posture is used as training data, the PDR AI model becomes capable of estimating the position in a normal posture, thereby overcoming the limitations of traditional PDR technology. Generally, even when performing gait detection using a PDR AI model, walking patterns vary slightly depending on the pedestrian, and there may be differences in sensor information patterns depending on the characteristics of the portable electronic device. Therefore, the accuracy of the PDR AI model may vary depending on the characteristics of the pedestrian or the portable electronic device. To address this, retraining must be performed by collecting training data from various users; however, conventional PDR AI model technology faces d