키넥트 센서 데이터 기반 보행분석 Ground-Truth 데이터 모델링

장용준 대한전자공학회 2022 전자공학회논문지 Vol.59 No.8

보행 분석은 사람 식별, 소아마비, 척수손상, 뇌졸중 환자 등의 운동 평가 및 재활, 퇴행성 뇌질환 진단 등 다양한 분야에서 활용되고 있다. 보행 분석에 있어서 모션 캡쳐의 Ground-Truth 데이터는 광학식 방식을 이용한 Vicon 또는 Optotrak 시스템 등으로부터 산출된 데이터이나, 이들 시스템은 마커를 부착해야하는 점과, 설치 공간상의 제약 및 상당한 고가의 시스템이라는 단점을 갖고 있다. 최근 보행 분석에 있어서 이러한 단점을 보완하고자 Microsoft 사의 Kinect 센서 여러개로 구성된 시스템으로부터 3D 공간에서 body point 위치 데이터를 추출하는 연구가 진행 중이다. 그러나 기존 연구는 제안 시스템의 유효성 평가에 있어서 Ground-Truth 데이터와 제안 시스템 산출 데이터 간의 선형상관관계, 재현가능성, 일치성 등을 증명할 뿐, Kinect 데이터를 기반으로 Ground-Truth 데이터를 모델링하고 그 값을 예측하는 연구는 아직 미비한 상황이다. 본 연구에서는 다변수 다중회귀분석을 이용하여 다중 Kinect 시스템으로부터 산출된 데이터를 기반으로 Ground-Truth 데이터를 모델링하고 예측하는 방법을 고찰하였고, 모델값과 실제 Ground-Truth 데이터값 간의 결정계수와 급내상관계수를 계산함으로써 상관관계에 대하여 분석하였다.

우량계 강우를 이용한 레이더 강우의 Ground-Truth 방법 검토

유철상,김병수,김경준,최정호 한국수자원학회 2007 한국수자원학회논문집 Vol.40 No.9

본 연구에서는 레이더 강우의 ground-truth를 위해 가능한 3가지 비교방법, 즉 3가지 ground-truth설계를 이론적으로 검토하였다. 이론적인 결과는 먼저, WGR강우모형으로 모의된 3차원 강우장에 적용하여 검토하였으며, 아울러 실측자료인 관악산 레이더 자료 및 레이더 관측영역 내 건설교통부 관할 지상관측소 강우 자료에 적용하여 평가하였다. 전체적인 적용결과는 이론적인 분석내용 및 모의 강우를 이용하여 수행한 내용과 유사함을 확인하였다. 결 This study theoretically compared three possible methods for the ground-truth, that is three ground-truth designs of radar rainfall using the rain gauge rainfall. Theoretical results derived are first applied to the rainfall field generated by the Waymire

Construction of a Video Dataset for Face Tracking Benchmarking Using a Ground Truth Generation Tool

Luu Ngoc Do,Hyung Jeong Yang,Soo Hyung Kim,Guee Sang Lee,In Seop Na,Sun Hee Kim 한국콘텐츠학회(IJOC) 2014 International Journal of Contents Vol.10 No.1

In the current generation of smart mobile devices, object tracking is one of the most important research topics for computer vision. Because human face tracking can be widely used for many applications, collecting a dataset of face videos is necessary for evaluating the performance of a tracker and for comparing different approaches. Unfortunately, the well-known benchmark datasets of face videos are not sufficiently diverse. As a result, it is difficult to compare the accuracy between different tracking algorithms in various conditions, namely illumination, background complexity, and subject movement. In this paper, we propose a new dataset that includes 91 face video clips that were recorded in different conditions. We also provide a semi-automatic ground-truth generation tool that can easily be used to evaluate the performance of face tracking systems. This tool helps to maintain the consistency of the definitions for the ground-truth in each frame. The resulting video data set is used to evaluate well-known approaches and test their efficiency.

Construction of a Video Dataset for Face Tracking Benchmarking Using a Ground Truth Generation Tool

Do, Luu Ngoc,Yang, Hyung Jeong,Kim, Soo Hyung,Lee, Guee Sang,Na, In Seop,Kim, Sun Hee The Korea Contents Association 2014 International Journal of Contents Vol.10 No.1

In the current generation of smart mobile devices, object tracking is one of the most important research topics for computer vision. Because human face tracking can be widely used for many applications, collecting a dataset of face videos is necessary for evaluating the performance of a tracker and for comparing different approaches. Unfortunately, the well-known benchmark datasets of face videos are not sufficiently diverse. As a result, it is difficult to compare the accuracy between different tracking algorithms in various conditions, namely illumination, background complexity, and subject movement. In this paper, we propose a new dataset that includes 91 face video clips that were recorded in different conditions. We also provide a semi-automatic ground-truth generation tool that can easily be used to evaluate the performance of face tracking systems. This tool helps to maintain the consistency of the definitions for the ground-truth in each frame. The resulting video data set is used to evaluate well-known approaches and test their efficiency.

LDA 및 BERTopic 기반 해외건설시장 뉴스 기사 토픽모델링 성능평가

백준우,정세환,지석호 대한토목학회 2023 대한토목학회논문집 Vol.43 No.6

해외건설사업 시, 현지 상황을 정확하고 빠르게 파악하는 것은 프로젝트 성공을 위해 매우 중요한 요소이다. 이는 토픽모델링을 활용한 뉴스 기사 분석을 통해 실현될 수 있다. 본 연구는 Latent Dirichlet Allocation(LDA)과 BERTopic 두 토픽모델링 기법을 활용하여 뉴스 기사를 분석하고, 최적의 기법을 찾고자 하였다. 모델링 결과로 자동생성된 토픽과 실제 문서 주제와의 일치 여부를 확인하기 위해 BBC 뉴스 기사 6,273건을 수집하여 ground truth를 생성하고, 이를 모델링된 토픽과 비교하였다. 그 결과 LDA의 F1 score는 0.011, BERTopic은 0.244로 나타났다. 이를 통해 BERTopic이 실제 뉴스 기사의 주제를 잘 파악하며, 해외건설시장의 주요 이슈를 자동으로 이해하는 데 더욱 용이하다는 것을 확인할 수 있었다. Understanding the local conditions is a crucial factor in enhancing the success potential of overseas construction projects. This can be achieved through the analysis of news articles of the target market using topic modeling techniques. In this study, the authors aimed to analyze news articles using two topic modeling methods, namely Latent Dirichlet Allocation (LDA) and BERTopic, in order to determine the optimal approach for market condition analysis. To evaluate the alignment between the generated topics and the actual themes of the news documents, the research collected 6,273 BBC news articles, created ground truth data for individual news article topics, and finally compared this ground truth with the results of the topic modeling. The F1 score for LDA was 0.011, while BERTopic achieved a score of 0.244. These results indicate that BERTopic more accurately reflected the actual topics of news articles, making it more effective for understanding the overseas construction market.

지능형 CCTV에서 다양한 객체 및 이벤트의 주석 데이터를 동적으로 저장하기 위한 DB 스키마 설계

안현석,김평근,권류혁,김경연,김유성 한국정보과학회 2014 데이타베이스 연구 Vol.30 No.1

To develop object and event recognition algorithms of intelligent surveillance systems, a lot of ground-truth data is essentially required. Therefore, many video annotation tools have been developed so far. However, the previous annotation tools have limitation to support reusing annotation data for other similar purposes while generating new annotation data for video is time-consuming task. If an annotation data can be reused for developing several other recognition algorithms, it is very helpful to reduce the development cost. Hence, in this paper, we propose a video annotation tool named INHA-VAS which is able to store the annotation data into the integrated annotation database instead of directly output ground truth data file. We also propose database schema for the integrated annotation database to accommodate diverse objects, events and their attributes for several different purposes. Hence, INHA-VAS can support well to generate annotation data for diverse objects and events and their attributes which is able to be reused as ground truth data for developing several different recognition algorithms for intelligent surveillance systems. 지능형 CCTV를 위한 객체 및 이벤트 인식 알고리즘을 개발하기 위해 대량의 지상 실측 데이터(ground-truthdata)가 필요하기 때문에 이를 효율적으로 생성하기 위한 비디오 주석 생성 시스템들이 개발되었다. 하지만 기존시스템들은 특정 객체 및 이벤트에 대해 생성한 주석 정보를 XML 문서나 텍스트 문서로 출력하기 때문에 이를다른 목적으로 사용하기 위해 부분 편집하는 것이 용이하지 않았다. 따라서 본 연구에서는 주석 정보를 통합 주석DB로 관리하도록 하고 다양한 목적에 따라 다른 구성내용으로 동적으로 지상 실측 데이터를 생성하여 지능형CCTV의 인식 알고리즘을 개발하는데 사용할 수 있도록 하였다. 즉, 통합 주석 DB의 스키마를 지능형 CCTV의사용 목적에 따라 다양한 객체 및 이벤트 그리고 각 객체 및 이벤트의 다양한 속성 정보를 관리 할 수 있도록설계하였다. 본 논문에서 제안하는 DB 스키마를 채택한 비디오 주석 생성 시스템(INHA-VAS)은 다양한 객체및 이벤트의 주석 정보를 저장할 수 있고 질의 검색기능을 이용하여 통합 주석 DB에서 필요한 정보만으로 XML형태의 지상 실측 데이터를 동적으로 생성할 수 있음을 예시하였다.

The necessity and availability of noise-free daily satellite-observed NDVI during rapid phenological changes in terrestrial ecosystems in East Asia

Shin Nagai,Taku M. Saitoh,Rikie Suzuki,Kenlo Nishida Nasahara,이우균,손요환,Hiroyuki Muraoka 한국산림과학회 2011 Forest Science And Technology Vol.7 No.4

General, global, long-term, and comprehensive phenological observations are required to evaluate the variability of photosynthetic activities due to environmental changes in terrestrial ecosystems. The observation of seasonal changes and detection of interannual variation in canopy phenology over regional and global scales require satellite data with high temporal resolution (i.e. a daily time step). However, satellite data often include noise caused by snow cover on vegetation, cloud contamination, and atmospheric aerosols. To accurately detect the timing of leaf-expansion and leaf-fall, which occur rapidly, and their rates, it is necessary to examine the observational frequency of noise-free satellite-observed vegetation index data during each phenological period. In this study, we investigated the spatiotemporal distribution of the number of observational days (NUMdays) in the Terra/MODIS (Moderate Resolution Imaging Spectroradiometer)-observed daily high-quality normalized difference vegetation index (NDVIhigh) data with no effects of snow cover, cloud contamination, or atmospheric noise. These data were examined for each month over 10 years in the various ecosystems of East Asia. To ground-truth the relationship between the Terra/MODISobserved daily NDVIhigh data and canopy surface images, we performed a long-term continuous field study in a cooltemperate deciduous broad-leaved forest in central Japan. During the leaf-expansion and leaf-fall periods, the NUMdays for NDVIhigh data in southern Russia, northeastern China, the Tibetan Plateau, Korea, and maritime Japan was about 3–7 for each month. The NUMdays for NDVIhigh data exceeded 10 for each month in arid regions during the growing season and in the subtropical region including northeastern India, Myanmar, and southwestern China during the dry season. In contrast, the NUMdays for NDVIhigh data was almost 0 for each month in southeastern China throughout the year and in the subtropical region during the southeastern monsoon season (July and August). By considering observations from both the Terra/MODIS and Aqua/MODIS satellites, the NUMdays for NDVIhigh data in the deciduous broad-leaved forest in Japan was increased by 40% compared with only Terra/MODIS satellite observations. Our findings indicate that daily NDVI data from multiple satellites detect the seasonal changes in the various ecosystems of East Asia more accurately than 8-day or biweekly composite NDVI data.

RGB-Depth 데이터 셋 개선을 통한 실내 환경의 단일 영상 깊이 추정

김지완,정민영,신영길 한국차세대컴퓨팅학회 2022 한국차세대컴퓨팅학회 논문지 Vol.18 No.4

Depth estimation from a single image is one of the fundamental task in computer vision. However, the depth images from the consumer-level depth camera still suffer from heavy noise and inaccurate qualities. Owing to the performances of supervised methods are fundamentally affected by the quality of the depth ground-truth (GT), constructing enhanced depth dataset is urged to improve the performances. In this study, we propose a framework to improve the performances of the supervised depth estimation methods based on a depth enhancement scheme. We trained the dataset on a baseline-level network, and performed comparable results with the state-of-the-art methods. Furthermore, the methods were retrained using the original and the enhanced dataset separately, and verified that the quality of GT dataset has primary role for the performance. 단일 영상에서의 깊이 추정은 컴퓨터 비전 분야에서 지속적으로 연구되어 오고 있는 분야이다. 그러나 상용화된 깊이 지각 카메라로부터 획득한 실내 영상의 깊이 맵은 여전히 부족한 정밀도와 상당한 노이즈를 포함하고 있다. 특히, 지도 학습 기반 깊이 추정 기법들의 성능은 정답 (ground-truth) 깊이 맵의 정밀도에 크게 영향을 받기 때문에 깊이 이미지의 품질 개선이 크게 요구된다. 본 논문에서는 개선된 실내 환경 RGB-Depth 데이터 셋을 통해 지도 학습 기반의 깊이 추정 기법들의 성능을 향상하는 방법을 제안한다. 실험을 통해 향상된 품질의 데이터 셋을 사용하여 단순한 네트워크 구조만으로도 최신의 방법들에 준하는 성능을 보였다. 또한, 기존의 데이터 셋과 제안하는 데이터 셋으로 각각 학습 시켰을 때 성능 향상을 보임으로써, 정답 데이터의 품질이 깊이 추정 성능에 주요한 역할을 수행한다는 것을 입증하였다.

Using SG Arrays for Hydrology in Comparison with GRACE Satellite Data, with Extension to Seismic and Volcanic Hazards

Crossley David,Hinderer Jacques The Korean Society of Remote Sensing 2005 大韓遠隔探査學會誌 Vol.21 No.1

We first review some history of the Global Geodynamics Project (GGP), particularly in the progress of ground-satellite gravity comparisons. The GGP Satellite Project has involved the measurement of ground-based superconducting gravimeters (SGs) in Europe for several years and we make quantitative comparisons with the latest satellite GRACE data and hydrological models. The primary goal is to recover information about seasonal hydrology cycles, and we find a good correlation at the microgal level between the data and modeling. One interesting feature of the data is low soil moisture resulting from the European heat wave in 2003. An issue with the ground-based stations is the possibility of mass variations in the soil above a station, and particularly for underground stations these have to be modeled precisely. Based on this work with a regional array, we estimate the effectiveness of future SG arrays to measure co-seismic deformation and silent-slip events. Finally we consider gravity surveys in volcanic areas, and predict the accuracy in modeling subsurface density variations over time periods from months to years.

Using SG Arrays for Hydrology in Comparison with GRACE Satellite Data, with Extension to Seismic and Volcanic Hazards

David Crossley,Jacques Hinderer 大韓遠隔探査學會 2005 大韓遠隔探査學會誌 Vol.21 No.1

We first review some history of the Global Geodynamics Project (GGP), particularly in the progress of ground-satellite gravity comparisons. The GGP Satellite Project has involved the measurement of ground-based superconducting gravimeters (SGs) in Europe for several years and we make quantitative comparisons with the latest satellite GRACE data and hydrological models. The primary goal is to recover information about seasonal hydrology cycles, and we find a good correlation at the microgal level between the data and modeling. One interesting feature of the data is low soil moisture resulting from the European heat wave in 2003. An issue with the ground-based stations is the possibility of mass variations in the soil above a station, and particularly for underground stations these have to be modeled precisely. Based on this work with a regional array, we estimate the effectiveness of future SG arrays to measure co-seismic deformation and silent-slip events. Finally we consider gravity surveys in volcanic areas, and predict the accuracy in modeling subsurface density variations over time periods from months to years.