• 대한임베디드공학회논문지
• /
• 제19권4호
• /
• pp.185-193
• /
• 2024

Visual-inertial odometry (VIO) is a method that leverages sensor data from a camera and an inertial measurement unit (IMU) for state estimation. Whereas conventional VIO has limited capability to estimate scale of translation, the performance of recent approaches has been improved by utilizing depth maps obtained from RGB-D camera, especially in indoor environments. However, the depth map obtained from the RGB-D camera tends to rapidly lose accuracy as the distance increases, and therefore, it is required to develop alternative method to improve the VIO performance in wide environments. In this paper, we argue that leveraging depth map estimated from a deep neural network has benefits to state estimation. To improve the reliability of depth information utilized in VIO algorithm, we propose a kernel-based sampling strategy to filter out depth values with low confidence. The proposed method aims to improve the robustness and accuracy of VIO algorithms by selectively utilizing reliable values of estimated depth maps. Experiments were conducted on real-world custom dataset acquired from underground parking lot environments. Experimental results demonstrate that the proposed method is effective to improve the performance of VIO, exhibiting potential for the use of depth estimation network for state estimation.

• Journal of Positioning, Navigation, and Timing
• /
• 제13권4호
• /
• pp.379-387
• /
• 2024

Real-time localization is essential for autonomous driving of robots. This paper presents the implementation and a performance analysis of a localization algorithm. To estimate the position and attitude of a robot, a visual inertial odometry (VIO) algorithm based on a multi-state constraint Kalman filter is used. The sensors employed in this study are a stereo camera and an inertial measurement unit (IMU). The performance is analyzed through experiments using three different camera view directions: floor-view, front-view, and ceiling-view. The number of detected features also affects navigation performance. Even if the number of recognized feature points is large, performance degrades if the correspondence between feature points is not accurately identified. The results show that VIO improves navigation performance even with low-cost sensors, thus facilitating map building as well as autonomous navigation.

• 전기학회논문지
• /
• 제68권1호
• /
• pp.189-198
• /
• 2019

In recent years, visual-inertial odometry(VIO) algorithms have been extensively studied for the indoor/urban environments because it is more robust to dynamic scenes and environment changes. In this paper, we propose loosely coupled(LC) VIO algorithm that utilizes the velocity vectors from both visual odometry(VO) and inertial measurement unit(IMU) as a filter measurement of Extended Kalman filter. Our approach improves the estimation performance of a filter without adding extra sensors while maintaining simple integration framework, which treats VO as a black box. For the VO algorithm, we employed a fundamental part of the ORB-SLAM, which uses ORB features. We performed an outdoor experiment using an RGB-D camera to evaluate the accuracy of the presented algorithm. Also, we evaluated our algorithm with the public dataset to compare with other visual navigation systems.

• 로봇학회논문지
• /
• 제18권3호
• /
• pp.260-270
• /
• 2023

We introduce a novel Visual Inertial Odometry (VIO) algorithm designed to improve the performance of thermal-inertial odometry. Thermal infrared image, though advantageous for feature extraction in low-light conditions, typically suffers from a high noise level and significant information loss during the 8-bit conversion. Our algorithm overcomes these limitations by approximating a 14-bit raw pixel histogram into a Gaussian mixture model. The conversion method effectively emphasizes image regions where texture for visual tracking is abundant while reduces unnecessary background information. We incorporate the robust learning-based feature extraction and matching methods, SuperPoint and SuperGlue, and zero velocity detection module to further reduce the uncertainty of visual odometry. Tested across various datasets, the proposed algorithm shows improved performance compared to other state-of-the-art VIO algorithms, paving the way for robust thermal-inertial odometry.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2009년도 추계학술대회 논문집
• /
• pp.617-618
• /
• 2009

• 한국게임학회 논문지
• /
• 제19권5호
• /
• pp.91-102
• /
• 2019

본 논문에서는 ARCore를 사용한 비전 기반 스마트폰용 AR 게임 개발을 하는데 있어 평면 추적의 정확도 개선 방법을 제안한다. ARCore에서는 SLAM과 스마트폰의 IMU를 혼합하여 사용하는 VIO 방식이다. IMU를 사용하는데 있어 가속도계와 자이로스코프의 단점으로 인해 한번 인식된 평면을 계속하여 추적하는데 있어 오차가 발생한다. 이를 위해 기존 센서뿐만 아니라 자기장계 센서를 추가하여 자세 측정의 오차를 최소화하고 이를 평면 추적에도 활용될 수 있도록 하는 방법을 제안한다. 제안 방법을 통해 평면 추적에 있어서 ARCore의 경우보다 스마트폰 자세 추정의 오차를 줄일 수 있는 것을 확인하였다.

• 한국정밀공학회:학술대회논문집
• /
• 한국정밀공학회 2009년도 추계학술대회 논문집
• /
• pp.31-32
• /
• 2009

• 한국정보통신학회논문지
• /
• 제13권7호
• /
• pp.1371-1378
• /
• 2009

본 논문에서는 로직 공정 기반의 저전력 eFuse OTP 메모리 셀을 제안하였다. eFuse OTP 메모리 셀은 프로그램과 읽기 모드에 최적화되도록 각각의 트랜지스터를 사용하였으며, WL과 BL의 기생적인 커패시턴스를 줄이므로 읽기 모드에서의 동작 전류를 줄였다. 그리고 저전력, 저면적의 eFuse OTP 메모리 IP 설계를 위하여 비동기식 인터페이스, 분리된 I/O, 디지털 센싱 방식의 BL 감지 증폭기 회로를 사용하였다. 모의실험 결과 읽기 모드에서의 동작전류는 VDD, VIO 각각 349.5${\mu}$A, 3.3${\mu}$A로 나왔다. 그리고 동부하이텍 0.18${\mu}$m generic 공정으로 설계된 eFuse OTP 메모 리 IP의 레이아웃 면적은300 ${\times}$557${\mu}m^2$이다.

• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송∙미디어공학회 2021년도 하계학술대회
• /
• pp.219-222
• /
• 2021

본 논문에서는 시각-관성 측위시스템(Visual-Inertial Odometry, VIO)에서 광학 문자 인식(Optical Character Recognition, OCR)을 활용해 문자의 영역을 찾아내고, 그 위치를 기억해 측위시스템에서 다시 인식되었을 때 비교하기 위해 위치와 특징점을 저장하고자 한다. 먼저, 실시간으로 움직이는 카메라의 영상에서 문자를 찾아내고, 카메라의 상대적인 위치를 이용하여 문자가 인식된 위치와 특징점을 저장하는 방법을 제안한다. 또한 저장된 문자가 다시 탐색되었을 때, 문자가 재인식되었는 지 판별하기 위한 방법을 제안한다. 인공적인 마커나 미리 학습된 객체를 사용하지 않고 상황에 따른 문자를 사용하는 이 방법은 문자가 존재하는 범용적인 공간에서 사용이 가능하다.

• 한국미생물·생명공학회지
• /
• 제52권2호
• /
• pp.215-217
• /
• 2024

Purple pigment producing bacterium strains AMJK, AMJM, and AMRM were isolated from sediment in sinan-gun, Korea and their draft genomes were sequenced using Illumina Hiseq 4000 platform. The lengths of AMJK, AMJM, and AMRM genomes were 6,380,747 bp, 6,381,259 bp, and 6,380,870 bp, respectively and G+C contents were 62.82%, 64.15%, and 62.82%, respectively. Comparative analysis of genomic identity showed that three strains were closely related to the group of Janthinobacterium lividum. Functional analysis of AMJK, AMJM, and AMRM genomes showed that all strains harbor genes related to producing violacein (VioABCDE).