• Journal of the Korea Society of Computer and Information
• /
• v.29 no.10
• /
• pp.129-135
• /
• 2024

In this paper, we design and implement a worker musculoskeletal assessment platform. The three core components of this platform are the Mobile App, the Modeling Server, and the Web Platform. The Mobile App is an Android application developed in Kotlin, targeting Android platform 12 (S) and Android API Level 31 devices. The app utilizes the camera to capture various worker motion data and transmits it to the Modeling Server. The Modeling Server is implemented using Node.js. This server converts the worker's motion data-such as points, skeleton, and x, y, z coordinate data, measured by the mobile app-into multidimensional arrays. It then applies machine learning frameworks like TensorFlow and Keras to predict the worker's posture. The worker posture learning model is built using Teachable Machine. The Web Platform is developed using React and visualizes the worker's movements as 3D animations along a timeline. The machine learning-based worker musculoskeletal assessment platform developed in this paper aims to contribute to minimizing musculoskeletal disorders in workers at industrial sites.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2023.07a
• /
• pp.513-515
• /
• 2023

본 논문에서는 그리드 탐색과 트랜스포머를 사용한 그룹 행동 인식 모델을 제안한다. 추출된 여러 사람의 스켈레톤 정보를 차분 벡터, 변위 벡터, 관계 벡터로 변환하고 사람별로 묶어 이를 TimeDistributed 함수에 넣고 풀링을 한다. 이를 트랜스포머 모델의 입력으로 넣고 그룹 행동 인식 분류를 출력하였다. 논문에서 3가지 벡터를 입력으로 하여 합치고 트랜스포머 계층을 거친 모델과 3가지 벡터를 입력으로 하고 계층적으로 트랜스포머 모델을 거쳐 행동 인식 분류를 출력하는 두 가지 모델을 제안한다. 3가지 벡터를 합친 모델에서 클래스 분류 정확도는 CAD 데이터 세트 96.6%, Volleyball 데이터 세트 91.4%, 계층적 트랜스포머 모델은 CAD 데이터 세트 96.8%, Volleyball 데이터 세트 91.1%를 얻었다

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.156-157
• /
• 2022

본 논문에서는 선택적 시점에서의 2D 포즈 추정(pose estimation) 결과를 정합 하여 정확도 높은 3D 스켈레톤(skeleton)을 만들어 낸다. 여러 프레임의 3D 데이터를 10 도 간격으로 36 방향에서 투영한 뒤, 2D 포즈 추정 결과 신뢰도가 높은 시점에서의 결과만을 선별하여 3 차원으로 정합 한다. 이때 사용하는 시점의 개수를 달리하며 정확도에 미치는 영향을 분석하여 실험적으로 정확도가 높은 최소의 시점 개수를 정하였다. 또한, 정합 한 3D 뼈대를 모션 캡쳐(motion capture) 센서와 비교하여 제안하는 알고리즘에 의해 3D 포즈 추정의 정확도가 향상되는 것을 확인했다.

• Annual Conference of KIPS
• /
• 2023.11a
• /
• pp.287-288
• /
• 2023

우리 사회가 고령화시대로 접어들면서 낙상은 매우 심각한 사회문제가 되고 있으며 정확한 낙상 감지 기술의 수요도 늘고 있다. 본 연구는 웹 캠을 이용한 개선된 낙상감지 기법을 제안한다. 제안하는 기법은 RGB 영상을 기반으로 스켈레톤 포즈 추출, 데이터 가공, GRU(Gated Recurrent Unit) 신경망 알고리즘을 적용한 낙상 감지 실험 및 감지 결과 분석의 과정이 포함된다.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.9
• /
• pp.109-114
• /
• 2024

In this paper, we design and implement a worker motion 3D visualization module based on human sensors. The three key modules that make up this system are Human Sensor Implementation, Data Set Creation, and Visualization. Human Sensor Implementation provides the functions of setting and installing the human sensor locations and collecting worker motion data through the human sensors. Data Set Creation offers functions for converting and storing motion data, creating near real-time worker motion data sets, and processing and managing sensor and motion data sets. Visualization provides functions for visualizing the worker's 3D model, evaluating motions, calculating loads, and managing large-scale data. In worker 3D model visualization, motion data sets (Skeleton & Position) are synchronized and mapped to the worker's 3D model, and the worker's 3D model motion animation is visualized by combining the worker's 3D model with analysis results. The human sensor-based worker motion 3D visualization module designed and implemented in this paper can be widely utilized as a foundational technology in the smart factory field in the future.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.3
• /
• pp.43-50
• /
• 2021

In this paper, we design and implement KITNESS, a windows application that feeds back the accuracy of fitness motions based on Kinect sensors. The feature of this application is to use Kinect's camera and joint recognition sensor to give feedback to the user to exercise in the correct fitness position. At this time, the distance between the user and the Kinect is measured using Kinect's IR Emitter and IR Depth Sensor, and the joint, which is the user's joint position, and the Skeleton data of each joint are measured. Using this data, a certain distance is calculated for each joint position and posture of the user, and the accuracy of the posture is determined. And it is implemented so that users can check their posture through Kinect's RGB camera. That is, if the user's posture is correct, the skeleton information is displayed as a green line, and if it is not correct, the inaccurate part is displayed as a red line to inform intuitively. Through this application, the user receives feedback on the accuracy of the exercise position, so he can exercise himself in the correct position. This application classifies the exercise area into three areas: neck, waist, and leg, and increases the recognition rate of Kinect by excluding positions that Kinect does not recognize due to overlapping joints in the position of each exercise area. And at the end of the application, the last exercise is shown as an image for 5 seconds to inspire a sense of accomplishment and to continuously exercise.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.23 no.6
• /
• pp.89-95
• /
• 2023

In recent decades, human action recognition (HAR) has demonstrated potential applications in sports analysis, human-robot interaction, and large-scale signage content. In this paper, spatial temporal attention graph convolutional network (STAGCN)-based HAR system is proposed. Spatioal-temmporal features of skeleton sequences are assigned different weights by STAGCN, enabling the consideration of key joints and viewpoints. From simulation results, it has been shown that the performance of the proposed model can be improved in terms of classification accuracy in the NTU RGB+D dataset.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.7
• /
• pp.279-286
• /
• 2021

Recently, We are carrying out a policy of physical distancing of at least 1m from each other to prevent the spreading of COVID-19 disease in public places. In this paper, we propose a method for measuring distances between people in real time and an automation system that recognizes objects that are within 1 meter of each other from stereo images acquired by drones or CCTVs according to the estimated distance. A problem with existing methods used to estimate distances between multiple objects is that they do not obtain three-dimensional information of objects using only one CCTV. his is because three-dimensional information is necessary to measure distances between people when they are right next to each other or overlap in two dimensional image. Furthermore, they use only the Bounding Box information to obtain the exact coordinates of human existence. Therefore, in this paper, to obtain the exact two-dimensional coordinate value in which a person exists, we extract a person's key point to detect the location, convert it to a three-dimensional coordinate value using Stereo Vision and Camera Calibration, and estimate the Euclidean distance between people. As a result of performing an experiment for estimating the accuracy of 3D coordinates and the distance between objects (persons), the average error within 0.098m was shown in the estimation of the distance between multiple people within 1m.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.9
• /
• pp.53-62
• /
• 2020

Product recommendation services that have been researched recently are only recommended through the customer's product purchase history. In this paper, we propose the digital signage service through customers' behavior pattern analysis that is recommending through not only purchase history, but also behavior pattern that customers take when choosing products. This service analyzes customer behavior patterns and extracts interests about products that are of practical interest. The service is learning extracted interest rate and customers' purchase history through the Wide & Deep model. Based on this learning method, the sparse vector of other products is predicted through the MF(Matrix Factorization). After derive the ranking of predicted product interest rate, this service uses the indoor signage that can interact with customers to expose the suitable advertisements. Through this proposed service, not only online, but also in an offline environment, it would be possible to grasp customers' interest information. Also, it will create a satisfactory purchasing environment by providing suitable advertisements to customers, not advertisements that advertisers randomly expose.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.20 no.6
• /
• pp.1183-1188
• /
• 2016

Kinect sensor version 2 is a kind of camera released by Microsoft as a computer vision and a natural user interface for game consoles like Xbox one. It allows acquiring color images, depth images, audio input and skeletal data with a high frame rate. In this paper, using depth image, we present a surveillance system of a certain area within Kinect's field of view. With computer vision library(Emgu CV), if an object is detected in the target area, it is tracked and kinect camera takes RGB image to send it in database server. Therefore, a mobile application on android platform was developed in order to notify the user that Kinect has sensed strange motion in the target region and display the RGB image of the scene. User gets the notification in real-time to react in the best way in the case of valuable things in monitored area or other cases related to a reserved zone.