• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.8
• /
• pp.3011-3024
• /
• 2021

Pose estimation of the sensor is important issue in many applications such as robotics, navigation, tracking, and Augmented Reality. This paper proposes visual-inertial integration system appropriate for dynamically moving condition of the sensor. The orientation estimated from Inertial Measurement Unit (IMU) sensor is used to calculate the essential matrix based on the intrinsic parameters of the camera. Using the epipolar geometry, the outliers of the feature point matching are eliminated in the image sequences. The pose of the sensor can be obtained from the feature point matching. The use of IMU sensor can help initially eliminate erroneous point matches in the image of dynamic scene. After the outliers are removed from the feature points, these selected feature points matching relations are used to calculate the precise fundamental matrix. Finally, with the feature point matching relation, the pose of the sensor is estimated. The proposed procedure was implemented and tested, comparing with the existing methods. Experimental results have shown the effectiveness of the technique proposed in this paper.

• Journal of Korea Multimedia Society
• /
• v.22 no.5
• /
• pp.547-557
• /
• 2019

To extract the depth map from a single image, a number of CNN-based deep learning methods have been performed in recent research. In this study, the GAN structure of Pix2Pix is maintained. this model allows to converge well, because it has the structure of the generator and the discriminator. But the convolution in this model takes a long time to compute. So we change the convolution form in the generator to a depthwise convolution to improve the speed while preserving the result. Thus, the seven down-sizing convolutional hidden layers in the generator U-Net are changed to depthwise convolution. This type of convolution decreases the number of parameters, and also speeds up computation time. The proposed model shows similar depth map prediction results as in the case of the existing structure, and the computation time in case of a inference is decreased by 64%.

• Journal of Auto-vehicle Safety Association
• /
• v.14 no.3
• /
• pp.71-76
• /
• 2022

In autonomous driving cars, 3D pose estimation can be one of the effective methods to enhance safety control for OOP (Out of Position) passengers. There have been many studies on human pose estimation using a camera. Previous methods, however, have limitations in automotive applications. Due to unexplainable failures, CNN methods are unreliable, and other methods perform poorly. This paper proposes robust real-time multi-human 3D pose estimation architecture in vehicle using monocular RGB camera. Using particle filter, our approach integrates CNN 2D/3D pose measurements with available information in vehicle. Computer simulations were performed to confirm the accuracy and robustness of the proposed algorithm.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.637-640
• /
• 2022

3차원 물체의 탐지와 자세 추정은 실내외 환경에서 장면 이해, 로봇의 물체 조작 작업, 자율 주행, 증강 현실 등과 같은 다양한 응용 분야들에서 공통적으로 요구되는 매우 중요한 시각 인식 기술이다. 깊이 지도를 요구하는 기존 연구들과는 달리, 본 논문에서는 RGB 컬러 영상만을 이용해 미지의 물체들, 즉 3차원 CAD 모델을 가지고 있지 않은 새로운 물체들을 탐지해내고, 이들의 자세를 추정해낼 수 있는 새로운 신경망 모델을 제안한다. 제안 모델에서는 최근 빠른 속도로 발전하고 있는 깊이 추정 기술을 이용함으로써, 깊이 측정 센서 없이도 물체 자세 추정에 필요한 깊이 지도를 컬러 영상에서 구해낼 수 있다. 본 논문에서는 벤치마크 데이터 집합을 이용한 실험을 통해, 제안 모델의 유용성을 평가한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.212-213
• /
• 2022

내시경 검사는 내장기관의 이상 유무를 점검할 수 있는 효과적인 의료 기술이다. 해당 논문에서는 자가지도 방식의 직접 감쇠 모델(DAM, Direct Attenuation Model)[3]을 사용한 내시경 비디오 기반 깊이 예측을 제안한다. 단안 카메라의 비디오 영상에서 DAM 을 이용한 빛의 밝기에 따른 깊이 변화 정보와 Normal 정보를 사용하여 깊이와 자세 예측 네트워크 모델 학습을 효과적으로 수행한다. 실험을 통해 제안하는 방법은 기존의 깊이 추정 네트워크 대비 다양한 내시경 비디오 영상에서 더 정확하게 깊이를 추정함을 확인하였다.

• Journal of the Korean Society for Precision Engineering
• /
• v.12 no.8
• /
• pp.156-167
• /
• 1995

In this paper, we propose a 3-D range finding method for situation that stereo camera has small translational motion. Binocular stereo generally tends to produce stereo correspondence errors and needs huge amount of computation. The former drawback is because the additional constraints to regularize the correspondence problem are not always true for every scene. The latter drawback is because they use either correlation or optimization to find correct disparity. We present a method which overcomes these drawbacks by moving the stereo camera actively. The method utilized a motion parallax acquired by monocular motion stereo to restrict the search range of binocular disparity. Using only the uniqueness of disparity makes it possible to find reliable binocular disparity. Experimental results with real scene are presented to demonstrate the effectiveness of this method.

• The Journal of Korea Robotics Society
• /
• v.18 no.1
• /
• pp.59-65
• /
• 2023

This paper proposes an Extended and Adaptive Inverse Perspective Mapping (EA-IPM) model that can obtain an accurate bird's-eye view (BEV) from the forward-looking monocular camera on the sidewalk with various curves. While Inverse Perspective Mapping (IPM) is a good way to obtain ground information, conventional methods assume a fixed relationship between the camera and the ground. Due to the nature of the driving environment of the mobile robot, there are more walking environments with frequent motion changes than flat roads, which have a fatal effect on IPM results. Therefore, we have developed an extended IPM process to be applicable in IPM on sidewalks by adding a formula for complementary Y-derive processes and roll motions to the existing adaptive IPM model that is robust to pitch motions. To convince the performance of the proposed method, we evaluated our results on both synthetic and real road and sidewalk datasets.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.95-96
• /
• 2021

본 논문에서는 단안비디오 입력으로부터 각 SAI(sub-aperture image)간의 넓은 기준선을 갖는 라이트필드 합성기법을 제안한다. 기존의 라이트필드 영상은 취득의 어려움에 의해 규모가 작고 특정 물체위주로 구성되어 있어 컴퓨터 비전 및 그래픽스 분야의 최신 딥러닝 기법들을 라이트필드 분야에 적용하기 어렵다는 문제를 갖고 있다. 이러한 문제점들을 해결하기 위해 사실적 렌더링 기반의 가상환경상에서 실제환경과 유사함을 갖는 데이터를 취득하였다. 생성한 데이터셋을 이용하여 기존의 새로운 시점을 생성하는 기법 중 하나인 다중 평면 영상(Multi Plane Image) 기반 합성기법을 통해 라이트필드 영상을 합성한다. 제안하는 네트워크는 단안비디오의 연속된 두개의 프레임으로부터 MPI 추정하는 네트워크와 입력영상의 깊이 정보를 추정하는 네트워크로 구성되어 있다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2021.06a
• /
• pp.322-323
• /
• 2021

증강현실이나 자율 주행, 드론 등의 기술에서 현재 위치와 시점을 파악하기 위해서는 실시간 카메라 자세 추정이 필요하다. 이를 위해 가장 일반적인 방식인 연속적인 단안 영상으로부터 카메라 자세를 추정하는 방식은 두 영상의 정적 객체 간에 견고한 특징점 매칭이 이루어져야한다. 하지만 일반적인 영상들은 다양한 이동 객체가 존재하는 동적 환경이므로 정적 객체만의 매칭을 보장하기 어렵다는 문제가 있다. 본 논문은 이 같은 동적 환경 문제를 해결하기 위해, 신경망 기반의 객체 분할 기법으로 영상 속 객체를 추출하고, 객체별 특징점 매칭 및 자세 추정 결과로 정적 객체를 특정해 매칭하는 방법을 제안한다. 또한, 제안하는 정적 객체 특정 방식에 적합한 신경망 기반 특징점 추출 방법을 사용하면 동적 환경에 보다 강인한 카메라 자세 추정이 가능함을 실험을 통해 확인한다.

• Journal of Korean Ophthalmic Optics Society
• /
• v.20 no.3
• /
• pp.385-390
• /
• 2015

Purpose: This study evaluated the static stereoacuity by Distance Randot Stereotest (STEREO OPTICAL. Co., Inc. USA) and the dynamic stereoacuity by three-rods test (iNT, Korea). Criterion and correlation of stereoacuity between both tests and usefulness of two stereotest methods were also evaluated. Methods: For normal adults of 109 (male 61, female 48), mean age of 20.88 (19-32 years) years old, static stereoacuity by using Distance Randot Stereotest at 3 m distance, dynamic stereoacuity by using three-rods test at 2.5 m distance were measured. Results: The mean of distance static stereoacuity was $155.77{\pm}133.11sec$ of arc and the mean of error distance dynamic stereoacuity $11.13{\pm}9.69mm$. With equivalent-conversion stereoacuity of $23.44{\pm}20.96sec$ of arc, there was statistically significant differences (p=0.00) between two dynamic stereoacuity, but correlation was relatively low (${\rho}=0.226$). In the case of dynamic stereoacuity, separated to normal range by criterion of the error distance 20 mm, it showed the error distance of less than 20 mm in 97 subjects(89%) whose average of error distance and conversion mean dynamic stereoacuity were $8.43{\pm}5.10mm$ and $17.68{\pm}10.67sec$ of arc. repectively. The error distance of was equivalent-conversion dynamic stereoacuity 40.99 sec of arc (PD 62 mm basis) was 20 mm. Conclusions: The results of lower correlation between static and dynamic stereoacuity suggest that seterotest should be applied separately to different functions. The results of this study also suggest that Distance Randot Stereotest can be applied to static stereoacuity excluding monocular cues. Three-rods test can be applied to dynamic stereoacuity containing the response of the eye-hand coordination in the daily life of natural vision condition, including the monocular cues. These different approaches canprovide a criterion of the two stereoacuity and parallel use of the two tests would be useful. For dynamic stereoacuity by three-rods test, error distance 20 mm in a normal range of adults can be used as a criteria to get statistical meaning of the results.