• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.737-741
• /
• 2009

This paper addresses the factorization method to estimate the projective structure of a scene from feature (points) correspondences over images with occlusions. We propose both a column and a row space approaches to estimate the depth parameter using the subspace constraints. The projective depth parameters are estimated by maximizing projection onto the subspace based either on the Joint Projection matrix (JPM) or on the the Joint Structure matrix (JSM). We perform the maximization over significant observation and employ Tardif's Camera Basis Constraints (CBC) method for the matrix factorization, thus the missing data problem can be overcome. The depth estimation and the matrix factorization alternate until convergence is reached. Result of Experiments on both real and synthetic image sequences has confirmed the effectiveness of our proposed method.

• Proceedings of the IEEK Conference
• /
• 2002.06d
• /
• pp.291-294
• /
• 2002

A simple computer vision technology to measure the middle-ranged depth with mono camera and plain mirror is proposed The proposed system is structured wiか the rotating mirror in front of the fixed mono camera In contrast to the previous stereo vision system in which the disparity of the closer object is larger than that of the distant object, the pixel movement caused by the rotating mirror is bigger for the pixels of the distant object in the proposed system Being inspired by such feature in the proposed system the principle of the depth measurement based on the relation of the pixel movement and the distance of object have been investigated. Also, the factors to influence the precision of the measurement are analysed The benefits of the proposed system are low price and less chance of occlusion. The robustness for practical usage is an additional benefit of the proposed vision system.

• Proceedings of the IEEK Conference
• /
• 2000.07b
• /
• pp.799-802
• /
• 2000

There are many methods that extract the depth information based on the blurring ratio for object point in DFD(Depth from Defocus). However, it is often difficult to measure the depth of the object in two-dimensional images that was affected by various elements such as edges, textures, and etc. To solve the problem, new DFD method employing the texture classification with a neural network is proposed. This method extracts the feature of texture from an evaluation window in an image and classifies the texture class. Finally, It allocates the correspondent value for the blurring ratio. The experimental result shows that the method gives more accurate than the previous methods.

• Journal of Korea Multimedia Society
• /
• v.19 no.9
• /
• pp.1659-1668
• /
• 2016

Understanding 3D structure of scenes is of a great interest in various vision-related tasks. In this paper, we present a unified approach for estimating depth from a single monocular image. The key idea of our approach is to take advantages both of parametric learning and non-parametric sampling method. Using a parametric convolutional network, our approach learns the relation of various monocular cues, which make a coarse global prediction. We also leverage the local prediction to refine the global prediction. It is practically estimated in a non-parametric framework. The integration of local and global predictions is accomplished by concatenating the feature maps of the global prediction with those from local ones. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods both qualitatively and quantitatively.

• Journal of the Korea Computer Graphics Society
• /
• v.24 no.5
• /
• pp.1-10
• /
• 2018

In this paper, we present a non-photorealistic rendering technique that automatically delivers a stylized abstraction of a photograph with DoF(Depth of field). Our approach is a new filtering method that efficiently classifies DoF regions using RGB channels and automatically adjusts the color abstraction and extracted line quality based on this classification. This DoF-based filtering is simple, fast, and easy to implement and significantly improves the abstraction performance in terms of feature enhancement and stylization.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.154-156
• /
• 2021

Depth-wise separable convolution 은 컴퓨터 자원이 제한된 환경에서 기존의 standard convolution을 대체하는데 강력하고, 효과적인 대안으로 잘 알려져 있다.[1] MobileNetV2 에서는 Inverted residual block을 소개한다. 이는 depth-wise separable convolution으로 인해 생기는 손실, 즉 channel 간의 데이터를 조합해 새로운 feature를 만들어낼 기회를 잃어버릴 때, 이를 depth-wise separable convolution 양단에 point-wise convolution(1×1 convolution)을 사용함으로써 극복해낸 block이다.[1] 하지만 1×1 convolution은 채널 수에 의존적(dependent)인 특징을 갖고 있고, 따라서 결국 네트워크가 깊어지면 깊어질수록 효율적이고(efficient) 가벼운(light weight) 네트워크를 만드는데 병목 현상(bottleneck)을 일으키고 만다. 이 논문에서는 channel-wise squeeze and excitation block(CSE)을 통해 1×1 convolution을 부분적으로 대체하는 방법을 통해 이 병목 현상을 해결한다.

• ETRI Journal
• /
• v.45 no.4
• /
• pp.666-677
• /
• 2023

In the last few years, convolutional neural networks (CNNs) have demonstrated good performance while solving various computer vision problems. However, since CNNs exhibit high computational complexity, signal processing is performed on the server side. To reduce the computational complexity of CNNs for edge computing, a lightweight algorithm, such as a MobileNet, is proposed. Although MobileNet is lighter than other CNN models, it commonly achieves lower classification accuracy. Hence, to find a balance between complexity and accuracy, additional hyperparameters for adjusting the size of the model have recently been proposed. However, significantly increasing the number of parameters makes models dense and unsuitable for devices with limited computational resources. In this study, we propose a novel MobileNet architecture, in which the number of parameters is adaptively increased according to the importance of feature maps. We show that our proposed network achieves better classification accuracy with fewer parameters than the conventional MobileNet.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.4
• /
• pp.674-679
• /
• 2006

In this paper a gesture recognition system using 3D feature data is described. The system relies on a novel 3D sensor that generates a dense range mage of the scene. The main novelty of the proposed system, with respect to other 3D gesture recognition techniques, is the capability for robust recognition of complex hand postures such as those encountered in sign language alphabets. This is achieved by explicitly employing 3D hand features. Moreover, the proposed approach does not rely on colour information, and guarantees robust segmentation of the hand under various illumination conditions, and content of the scene. Several novel 3D image analysis algorithms are presented covering the complete processing chain: 3D image acquisition, arm segmentation, hand -forearm segmentation, hand pose estimation, 3D feature extraction, and gesture classification. The proposed system is tested in an application scenario involving the recognition of sign-language postures.

• Proceedings of the IEEK Conference
• /
• 2002.07c
• /
• pp.1404-1407
• /
• 2002

Although we look at objects as 2D images through our eyes, we can reconstruct the shape and/or depth of objects. In order to realize this ability using computers, it is required that the method which can estimate the 3D features of object from 2D images. As feature which represents 3D shapes effectively, three dimensional vector autoregressive model is pro- posed. If this feature is associated other feature of 2D shape, then above aim might be achieved. On the other hand, as feature which represents 2D shapes, quasi moment features is proposed. As the first step of association of these features, we constructed real time simulator that computes both of two features concurrently from object data (3D curves) . This simulator can also rotate object and estimate the rotation The method using 3D VAR model estimates the rotation correctly, but the estimation by quasi moment features includes much errors. This reason would be that projected images are constructed by the points only, and doesn't have enough sizes to estimate the correct 3D rotation parameters.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.13 no.10
• /
• pp.4807-4813
• /
• 2012

Since the existing methods of segmenting target objects from various images mainly use 2-dimensional features, they have several constraints due to the shortage of 3-dimensional information. In this paper, we therefore propose a new method of accurately segmenting target objects from three dimensional stereoscopic images using 2D and 3D feature clustering. The suggested method first estimates depth features from stereo images by using a stereo matching technique, which represent the distance between a camera and an object from left and right images. It then eliminates background areas and detects foreground areas, namely, target objects by effectively clustering depth and color features. To verify the performance of the proposed method, we have applied our approach to various stereoscopic images and found that it can accurately detect target objects compared to other existing 2-dimensional methods.