• Journal of Intelligence and Information Systems
• /
• v.25 no.1
• /
• pp.163-177
• /
• 2019

As smartphones are getting widely used, human activity recognition (HAR) tasks for recognizing personal activities of smartphone users with multimodal data have been actively studied recently. The research area is expanding from the recognition of the simple body movement of an individual user to the recognition of low-level behavior and high-level behavior. However, HAR tasks for recognizing interaction behavior with other people, such as whether the user is accompanying or communicating with someone else, have gotten less attention so far. And previous research for recognizing interaction behavior has usually depended on audio, Bluetooth, and Wi-Fi sensors, which are vulnerable to privacy issues and require much time to collect enough data. Whereas physical sensors including accelerometer, magnetic field and gyroscope sensors are less vulnerable to privacy issues and can collect a large amount of data within a short time. In this paper, a method for detecting accompanying status based on deep learning model by only using multimodal physical sensor data, such as an accelerometer, magnetic field and gyroscope, was proposed. The accompanying status was defined as a redefinition of a part of the user interaction behavior, including whether the user is accompanying with an acquaintance at a close distance and the user is actively communicating with the acquaintance. A framework based on convolutional neural networks (CNN) and long short-term memory (LSTM) recurrent networks for classifying accompanying and conversation was proposed. First, a data preprocessing method which consists of time synchronization of multimodal data from different physical sensors, data normalization and sequence data generation was introduced. We applied the nearest interpolation to synchronize the time of collected data from different sensors. Normalization was performed for each x, y, z axis value of the sensor data, and the sequence data was generated according to the sliding window method. Then, the sequence data became the input for CNN, where feature maps representing local dependencies of the original sequence are extracted. The CNN consisted of 3 convolutional layers and did not have a pooling layer to maintain the temporal information of the sequence data. Next, LSTM recurrent networks received the feature maps, learned long-term dependencies from them and extracted features. The LSTM recurrent networks consisted of two layers, each with 128 cells. Finally, the extracted features were used for classification by softmax classifier. The loss function of the model was cross entropy function and the weights of the model were randomly initialized on a normal distribution with an average of 0 and a standard deviation of 0.1. The model was trained using adaptive moment estimation (ADAM) optimization algorithm and the mini batch size was set to 128. We applied dropout to input values of the LSTM recurrent networks to prevent overfitting. The initial learning rate was set to 0.001, and it decreased exponentially by 0.99 at the end of each epoch training. An Android smartphone application was developed and released to collect data. We collected smartphone data for a total of 18 subjects. Using the data, the model classified accompanying and conversation by 98.74% and 98.83% accuracy each. Both the F1 score and accuracy of the model were higher than the F1 score and accuracy of the majority vote classifier, support vector machine, and deep recurrent neural network. In the future research, we will focus on more rigorous multimodal sensor data synchronization methods that minimize the time stamp differences. In addition, we will further study transfer learning method that enables transfer of trained models tailored to the training data to the evaluation data that follows a different distribution. It is expected that a model capable of exhibiting robust recognition performance against changes in data that is not considered in the model learning stage will be obtained.

• Journal of the Korean Institute of Gas
• /
• v.17 no.2
• /
• pp.28-35
• /
• 2013

The number of large and complex buildings is growing and they are usually concentrated in metropolitan cities. There is a possibility in such buildings that a small accident can expand to a massive disaster since their scale and complexity. To deal with this issue, a research on gas sensors which can detect multiple gases and early-warning systems has been conducted. Proper criteria or standards are necessary for effective application and operation of such sensor-based disaster monitoring system. In this study, we have proposed the alarm criteria of concentration of hazardous gases for the detection and the alarm release. For each alarm level, systematic disaster response plans consist of responsive actions and information delivery have been prepared. These disaster monitoring criteria can help the detection of hazardous gas-related disaster in the early stage of accident and the provision of appropriate emergency responses.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.5 no.1
• /
• pp.1-6
• /
• 2004

Air qualify monitoring if a primary activity for industrial and social environment. Especially, the VOCs(Volatile Organic Compounds) are very harmful for human and environment. Throughout this research. we designed sensor array with various kinds of gas sensor, and the recognition algorithm with ANN(Artificial Neural Network : BP), respectively. We have designed system to recognize various kinds and quantities of VOCs, such as benzene, tolylene, and xylene.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.589-591
• /
• 2023

현재 수치예보 시스템은 항공기, 위성 등 다양한 센서에서 얻은 다종 관측 데이터를 동화하여 대기 상태를 추정하고 있지만, 관측변수 또는 물리량이 서로 다른 관측들을 처리하기 위한 계산 복잡도가 매우 높다. 본 연구에서 기존 시스템의 계산 효율성을 개선하여 관측을 평가하거나 전처리하는 데에 효율적으로 활용하기 위해, 각 관측의 특성을 고려한 자기 지도학습 방법을 통해 멀티모달 기상관측으로부터 실제 대기 상태를 추정하는 방법론을 제안하고자 한다. 비균질적으로 수집되는 멀티모달 기상관측 데이터를 융합하기 위해, (i) 기상관측의 heterogeneous network를 구축하여 개별 관측의 위상정보를 표현하고, (ii) pretext task 기반의 self-supervised learning을 바탕으로 개별 관측의 특성을 표현한다. (iii) Graph neural network 기반의 예측 모델을 통해 실제에 가까운 대기 상태를 추정한다. 제안하는 모델은 대규모 수치 시뮬레이션 시스템으로 수행되는 기존 기술의 한계점을 개선함으로써, 이상 관측 탐지, 관측의 편차 보정, 관측영향 평가 등 관측 전처리 기술로 활용할 수 있다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2023.05a
• /
• pp.433-433
• /
• 2023

토양수분은 물 순환의 필수적인 요소로써 수문순환 및 기상 현상에 큰 영향을 미친다. 현재 우리나라에서는 토양수분 자료구축을 위해 Frequency Domain Reflectometry (FDR), Time Domain Reflectometry (TDR) 센서를 활용하여 지점 단위 토양수분 자료를 생산하고 있다. 그러나 한반도는 도서, 산간 지역이 다수 분포하고 있어, 지점관측 센서만으로 공간 대표성을 갖는 토양수분 자료를 산출하기 어렵다. 이에, 광범위한 지역을 장기간 모니터링 할 수 있는 원격탐사 기법을 활용하여, Advanced SCATterometer (ASCAT), Soil Moisture Active and Passive (SMAP) 등의 공간 단위 토양수분 자료의 적용성이 평가되고 있다. 하지만, 공간 토양수분 자료의 검증을 위해 필수적인 지점 토양수분 자료가 구축되지 않은 미계측지역이 다수 존재하며, 한반도와 같이 지형적 복잡성이 높게 나타나는 지역에서는 계측지역에서의 활용성 평가 결과가 미계측지역에서도 유사하게 나타난다고 가정하기 어렵다. 이에 본 연구에서는, 미계측지역의 공간 토양수분 자료를 산출하고자 계측지역에서 SM2RAIN 알고리즘으로 산출된 강수량 자료와 위성 산출 자료 그리고 지점관측 자료의 관계성을 분석했다. SM2RAIN 알고리즘의 입력자료는 Advanced SCATterometer (ASCAT) 토양수분 자료를 활용했다. ASCAT 토양수분 자료와 SM2RAIN 강수 자료의 검증을 위해 기상청에서 제공하는 Automated Agriculture Observing System (AAOS) 토양수분 자료, Automatic Weather System (AWS) 강수량 자료와 Global Precipitation Measurement (GPM) 강수 자료를 활용하였다. 전반적으로 ASCAT 토양수분을 통해 산출한 SM2RAIN 강수량의 추정과GPM 강수량이 유의미한 상관성이 나타나는 것을 확인할 수 있었으며, 추후 Downscaling 기법과 연계하여 지형적 복잡성이 높게 나타나는 지역의 토양수분 추정이 가능할 것으로 기대된다.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2022.05a
• /
• pp.439-439
• /
• 2022

인공위성은 위성통신, 기상 등 다양한 분야에서 활용되고 있지만 재난과 위성영상 특성 매칭의 제약으로 재난 상황에서는 제한적으로 사용되었다. 국내외 위성 갯수의 증가로 위성영상을 준-실시간으로 확보 가능함에 따라 활용할 수 있는 범위가 증가하여 최근에는 재난·재해에 신속하게 대비하기 위한 연구가 활발히 진행되고 있다. 본 연구는 재난 발생 지역의 위성 영상 확보를 위해 촬영된 영상과 미래시점의 촬영 예정인 영상의 촬영 예정 시간 및 영역을 빠른 시간 내 분석하여 최적 위성영상 확보에 기반이 되고자 한다. 행정안전부에서 분류한 재난·재해 유형에 따라 재난 예측, 탐지, 사후처리를 위한 위성자료의 확보를 위하여 다양한 위성과 탑재된 센서들의 궤도, 공간 해상도, 파장대 등의 위성영상의 적시성을 분석하여 최적 위성을 정의하였다. 위성 궤도 시뮬레이션은 TLE(Two Line Element) 정보를 이용하는 SGP4(Simplified General Perturbations version 4) 모델에 적용하여 개발하였다. 최신 TLE 정보를 이용하여 위성 궤도 정보 및 센서 정보(공간 해상도, Swath width, incidence angle IFOV 등)을 적용하였다. 수집된 위성 궤도 정보를 기반으로 위성의 궤도를 예측하여 예측된 위치에서의 촬영 영역을 산정하는 분석 기능을 수행하여 최종 시뮬레이션 데이터를 생성한다. 개발된 위성 궤도 시뮬레이션 알고리즘을 토대로 태풍 미탁 사례에 적용하였다. 위성 궤도 시뮬레이션 알고리즘을 태풍 미탁 사례에 적용한 결과 다종 위성리스트 중 위성 궤도 분석을 통해 최단기간 획득 가능한 위성 중 정지 궤도 기상위성인 Himawari-8, GK-2A는 태풍 경로 모니터링, 광학 위성인 Sentinel-2, PlanetScope는 건물 피해 지역, SAR 위성인 Sentinel-1, ICEYE는 홍수 지역을 탐지하는데 최적 위성 영상으로 분석되었다.

• The Sea:JOURNAL OF THE KOREAN SOCIETY OF OCEANOGRAPHY
• /
• v.19 no.3
• /
• pp.202-214
• /
• 2014

We introduce status and prospect of increasingly utilizing, unmanned, global ocean observing systems, and the global network to integrate, coordinate, and manage the systems. Platforms of the ocean observing system are diversified in order to resolve/monitor the variability occurring at multiple scales in both three-dimensional space and time. Here purpose, development history, and current status of the systems in two kinds - mobile (surface drifter, subsurface float, underwater glider) and fixed platforms (surface and subsurface moorings, bottom mounts), are examined and the increased future uses to produce synergies are envisioned. Simultaneous use of various mobile and fixed platforms is suggested to more effectively design the observing system, with an example of the NSF-funded OOI (Ocean Observations Initiative) program. Efforts are suggested 1) to fill the data gap existing in the deep sea and the Southern Ocean, and toward 2) new global network for oceanic boundary currents, 3) new technologies for existing and new sensors including biogeochemical, acoustic, and optical sensors, 3) data standardization, and 4) sensor calibration and data quality control.

• 전자공학회논문지 IE
• /
• v.45 no.3
• /
• pp.27-35
• /
• 2008

In this paper, we propose a minute signal detection algorithm for a development of optical analyzer, using the non-dispersive infrared method with multi gas filter correlation wheel, that can measure various environmental air-pollution materials (CO, SO2, NOx, etc.) in real-time. The MCT(mercury cadmium telluride) sensor can detect minute signals those show and absorption characteristic of each environmental pollution materials. In the proposed method, a corresponding data of each environmental pollution materials can be separated by an external trigger and threshold values in the measured continuous signals.

• Korean Journal of Remote Sensing
• /
• v.37 no.3
• /
• pp.543-557
• /
• 2021

Because aerosols have different spectral characteristics according to the size and composition of the particle and to the satellite sensors, a comparative analysis of aerosol products from various satellite sensors is required. In South Korea, however, a comprehensive study for the comparison of various official satellite AOD (Aerosol Optical Depth) products for a long period is not easily found. In this paper, we aimed to assess the performance of the AOD products from MODIS (Moderate Resolution Imaging Spectroradiometer), VIIRS (Visible Infrared Imaging Radiometer Suite), Himawari-8, and Sentinel-3 by referring to the AERONET (Aerosol Robotic Network) sun photometer observations for the period between January 2015 and December 2019. Seasonal and geographical characteristics of the accuracy of satellite AOD were also analyzed. The MODIS products, which were accumulated for a long time and optimized by the new MAIAC (Multiangle Implementation of Atmospheric Correction) algorithm, showed the best accuracy (CC=0.836) and were followed by the products from VIIRS and Himawari-8. On the other hand, Sentinel-3 AOD did not appear to have a good quality because it was recently launched and not sufficiently optimized yet, according to ESA (European Space Agency). The AOD of MODIS, VIIRS, and Himawari-8 did not show a significant difference in accuracy according to season and to urban vs. non-urban regions, but the mixed pixel problem was partly found in a few coastal regions. Because AOD is an essential component for atmospheric correction, the result of this study can be a reference to the future work for the atmospheric correction for the Korean CAS (Compact Advanced Satellite) series.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.14 no.12
• /
• pp.6420-6426
• /
• 2013

The indoor areas for the commercial use of automatic monitoring systems of mobile robot localization improves the cognitive abilities and the needs of the environment with this emerging and existing mobile robot localization, and object recognition methods commonly around its great sensor are leveraged. On the other hand, there is a difficulty with a problem-solving self-location estimation in indoor mobile robots using only the sensors of the robot. Therefore, in this paper, a self-position estimation method for an enhanced and effective mobile robot is proposed using a marker and CCTV video that is already installed in the building. In particular, after recognizing a square mobile robot and the object from the input image, and the vertices were confirmed, the feature points of the marker were found, and marker recognition was then performed. First, a self-position estimation of the mobile robot was performed according to the relationship of the image marker and a coordinate transformation was performed. In particular, the estimation was converted to an absolute coordinate value based on CCTV information, such as robots and obstacles. The study results can be used to make a convenient self-position estimation of the robot in the indoor areas to verify the self-position estimation method of the mobile robot. In addition, experimental operation was performed based on the actual robot system.