• The KIPS Transactions:PartB
• /
• v.12B no.3 s.99
• /
• pp.247-256
• /
• 2005

One of the classic research problems in computer vision is that of stereo, i.e., the reconstruction of three dimensional shape from two or more images. This paper deals with the problem of extracting depth information of non-rigid dynamic 3D scenes from general 2D video sequences taken by monocular camera, such as movies, documentaries, and dramas. Depth of the blocks are extracted from the resultant block motions throughout following two steps: (i) calculation of global parameters concerned with camera translations and focal length using the locations of blocks and their motions, (ii) calculation of each block depth relative to average image depth using the global parameters and the location of the block and its motion, Both singular and non-singular cases are experimented with various video sequences. The resultant relative depths and ego-motion object shapes are virtually identical to human vision.

• Journal of the Institute of Convergence Signal Processing
• /
• v.3 no.2
• /
• pp.7-14
• /
• 2002

This paper proposes a camera motion estimation technique using special relations of motion vectors of geometrically symmetrical triple points of two consecutive views of single camera. The proposed technique uses camera-induced motion vectors and their relations other than feature points and epioplar constraints. As contrast to the time consuming iterations or numerical methods in the calculation of E-matrix or F-matrix induced by epipolar constraints, the proposed technique calculates camera motion parameters such as panning, tilting, rolling, and zooming at once by applying the proposed linear equation sets to the motion vectors. And by devised background discriminants, it effectively reflects only the background region into the calculation of motion parameters, thus making the calculation more accurate and fast enough to accommodate MPEG-4 requirements. Experimental results on various types of sequences show the validity and the broad applicability of the proposed technique.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.37 no.12
• /
• pp.1445-1452
• /
• 2013

In this paper, we propose a real-time algorithm for estimating the relative position of a person with respect to a robot (camera) using a monocular camera. The algorithm detects the head and shoulder regions of a person using HOG (Histogram of Oriented Gradient) feature vectors and an SVM (Support Vector Machine) classifier. The size and location of the detected area are used for calculating the relative distance and angle between the person and the camera on a robot. To increase the speed of the algorithm, we use a GPU and NVIDIA's CUDA library; the resulting algorithm speed is ~ 15 Hz. The accuracy of the algorithm is compared with the output of a SICK laser scanner.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2022.10a
• /
• pp.256-257
• /
• 2022

In this paper, we introduce a method of extracting three-dimensional feature points through the movement of a single mobile device. Using a monocular camera, a 2D image is acquired according to the camera movement and a baseline is estimated. Perform stereo matching based on feature points. A feature point and a descriptor are acquired, and the feature point is matched. Using the matched feature points, the disparity is calculated and a depth value is generated. The 3D feature point is updated according to the camera movement. Finally, the feature point is reset at the time of scene change by using scene change detection. Through the above process, an average of 73.5% of additional storage space can be secured in the key point database. By applying the algorithm proposed to the depth ground truth value of the TUM Dataset and the RGB image, it was confirmed that the\re was an average distance difference of 26.88mm compared with the 3D feature point result.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.23 no.1
• /
• pp.123-133
• /
• 2024

Existing autonomous driving technology has been developed based on sensors attached to the vehicles to detect the environment and formulate driving plans. On the other hand, it has limitations, such as performance degradation in specific situations like adverse weather conditions, backlighting, and obstruction-induced occlusion. To address these issues, cooperative autonomous driving technology, which extends the perception range of autonomous vehicles through the support of road infrastructure, has attracted attention. Nevertheless, the real-time analysis of the 3D centroids of objects, as required by international standards, is challenging using single-lens cameras. This paper proposes an approach to detect objects and estimate the centroid of vehicles using the fixed field of view of road infrastructure and pre-measured geometric information in real-time. The proposed method has been confirmed to effectively estimate the center point of objects using GPS positioning equipment, and it is expected to contribute to the proliferation and adoption of cooperative autonomous driving infrastructure technology, applicable to both vehicles and road infrastructure.

• Journal of KIISE:Software and Applications
• /
• v.35 no.12
• /
• pp.780-788
• /
• 2008

Modeling hand poses and tracking its movement are one of the challenging problems in computer vision. There are two typical approaches for the reconstruction of hand poses in 3D, depending on the number of cameras from which images are captured. One is to capture images from multiple cameras or a stereo camera. The other is to capture images from a single camera. The former approach is relatively limited, because of the environmental constraints for setting up multiple cameras. In this paper we propose a method of reconstructing 3D hand poses from a 2D input image sequence captured from a single camera by means of Belief Propagation in a graphical model and recognizing a finger clicking motion using a hidden Markov model. We define a graphical model with hidden nodes representing joints of a hand, and observable nodes with the features extracted from a 2D input image sequence. To track hand poses in 3D, we use a Belief Propagation algorithm, which provides a robust and unified framework for inference in a graphical model. From the estimated 3D hand pose we extract the information for each finger's motion, which is then fed into a hidden Markov model. To recognize natural finger actions, we consider the movements of all the fingers to recognize a single finger's action. We applied the proposed method to a virtual keypad system and the result showed a high recognition rate of 94.66% with 300 test data.

• Journal of Broadcast Engineering
• /
• v.19 no.6
• /
• pp.942-956
• /
• 2014

This paper proposes a real-time hand pose tracking method using a monocular RGB camera. Hand tracking has high ambiguity since a hand has a number of degrees of freedom. Thus, to reduce the ambiguity the proposed method adopts the step-by-step estimation scheme: a palm pose estimation, a finger yaw motion estimation, and a finger pitch motion estimation, which are performed in consecutive order. Assuming a hand to be a plane, the proposed method utilizes a planar hand model, which facilitates a hand model regeneration. The hand model regeneration modifies the hand model to fit a current user's hand, and improves robustness and accuracy of the tracking results. The proposed method can work in real-time and does not require GPU-based processing. Thus, it can be applied to various platforms including mobile devices such as Google Glass. The effectiveness and performance of the proposed method will be verified through various experiments.

• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.341-350
• /
• 2022

In this paper, we propose a deep learning based on-device augmented reality (AR) system in which multiple input images are used to implement the correct occlusion in a real environment. The proposed system is composed of three technical steps; camera pose estimation, depth estimation, and object augmentation. Each step employs various mobile frameworks to optimize the processing on the on-device environment. Firstly, in the camera pose estimation stage, the massive computation involved in feature extraction is parallelized using OpenCL which is the GPU parallelization framework. Next, in depth estimation, monocular and multiple image-based depth image inference is accelerated using the mobile deep learning framework, i.e. TensorFlow Lite. Finally, object augmentation and occlusion handling are performed on the OpenGL ES mobile graphics framework. The proposed augmented reality system is implemented as an application in the Android environment. We evaluate the performance of the proposed system in terms of augmentation accuracy and the processing time in the mobile as well as PC environments.

• Journal of Aerospace System Engineering
• /
• v.17 no.6
• /
• pp.16-26
• /
• 2023

With the growing demand for unmanned aerial vehicles (UAVs), various collision avoidance methods have been proposed, mainly using LiDAR and stereo cameras. However, it is difficult to apply these sensors to small UAVs due to heavy weight or lack of space. The recently proposed methods use a combination of object recognition models and distance sensors, but they lack information on the obstacle size. This disadvantage makes distance determination and obstacle coordination complicated in an early-stage collision avoidance. We propose a method for estimating obstacle sizes using a monocular camera-YOLO and infrared sensor. Our experimental results confirmed that the accuracy was 86.39% within the distance of 40 cm. In addition, the proposed method was applied to a small UAV to confirm whether it was possible to avoid obstacle collisions.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.11a
• /
• pp.87-90
• /
• 2009

영상의 깊이 정보를 추출하는 것은 매우 어려운 연구이다. 다양한 유형의 영상 구조의 분석이 필요하지만 많은 경우에 주관적인 판단의 도움이 필요하다. 본 논문에서는 로스 텍스처 필터를 기반으로 정지 영상의 깊이를 자동으로 생성하는 방법을 제안한다. 로스 텍스처 필터는 단안 비전에서 3D 깊이를 얻기 위한 방법으로 활용되었는데, 실제 2D 영상에서 깊이를 예측하기 위해 텍스처 편차, 텍스처 기울기, 색상 등을 활용한다. 로스 필터는 $1{\times}5$ 벡터로부터 콘볼루션을 이용하여, 20여개의 $5{\times}5$ 콘볼루션 필터가 구해지는데, 영상에 필터를 적용하여 로스 에너지를 계산한다. 구해진 에너지를 깊이 맵으로 변환하고, 깊이 맵에서 특징 점을 구하고, 특징 점들로부터 델러노이 삼각화를 이용하여 삼각형 깊이 메쉬를 얻는다. 구해진 깊이 맵의 성능을 측정하기 위해 카메라 시점을 변경하면서 영상의 3D 구조를 분석하였으며, 입체영상을 생성하여 3D 입체 시청 결과를 분석하였다. 실험에서는 로스 텍스처 필터를 이용하는 깊이 생성 방법이 좋은 효과를 얻는 것을 확인하였다.