• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2023년도 추계학술발표대회
• /
• pp.557-560
• /
• 2023

In this paper, we introduce a pre-training method leveraging the capabilities of the Vision Transformer (ViT) for disease diagnosis in conventional Fundus images. Recognizing the need for effective representation learning in medical images, our method combines the Vision Transformer with a Masked Autoencoder to generate meaningful and pertinent image augmentations. During pre-training, the Masked Autoencoder produces an altered version of the original image, which serves as a positive pair. The Vision Transformer then employs contrastive learning techniques with this image pair to refine its weight parameters. Our experiments demonstrate that this dual-model approach harnesses the strengths of both the ViT and the Masked Autoencoder, resulting in robust and clinically relevant feature embeddings. Preliminary results suggest significant improvements in diagnostic accuracy, underscoring the potential of our methodology in enhancing automated disease diagnosis in fundus imaging.

• 대한인간공학회지
• /
• 제19권1호
• /
• pp.91-100
• /
• 2000

In this paper the intent recognition system which recognizes the human's head movements as a simple non-verbal intent representation is presented. The system recognizes five basic intent representations. i.e., strong/weak affirmation. strong/weak negation, and ambiguity by image processing of nodding or shaking movements of head. The vision system for tracking the head movements is composed of CCD camera, image processing board and personal computer. The modified template matching method which replaces the reference image with the searched target image in the previous step is used for the robust tracking of the head movements. For the improvement of the processing speed, the searching is performed in the pyramid representation of the original image. By inspecting the variance of the head movement trajectories. we can recognizes the two basic intent representations - affirmation and negation. Also, by focusing the speed of the head movements, we can see the possibility which recognizes the strength of the intent representation.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2003년도 학술회의 논문집 정보 및 제어부문 A
• /
• pp.231-234
• /
• 2003

This paper describes the structure, operation, image processing, and decision making techniques of a color machine vision system designed for the automatic sorting of soybeans. The system consists of feeder, conveyor belt, line-scan camera, lights. ejector, and a PC Unlike manufactured goods, agricultural products including soybeans have quite uneven features. The criteria for sorting good and bad beans also vary depending on inspectors. We tackle these problem by letting the system learn the inspecting parameters from good samples selected manually by a machine user before running the system for sorting. Real-time processing has another importance In the design. Four parallel DSPs are employed to increase the processing speed. When the designed system was tested with real soybeans and the result was successful.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1999년도 하계학술대회 논문집 B
• /
• pp.797-799
• /
• 1999

In this paper, the edge intensification and detection algorithm which is one of image processing operations is considered. Edge detection algorithm is the most useful and important method for image processing or image analysis. The vision system based on these processing and concerned in specific project is proposed and is applied to the inverted pendulum in order to automatically acquire the angles between the bar and the perpendicular reference line. In this paper, the angles that are obtained from some images of computer vision system can offer useful informations for control of real inverted pendulum system. Next, the inverted pendulum will be controlled by the proposed method.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1995년도 추계학술대회 논문집 학회본부
• /
• pp.587-589
• /
• 1995

The safe navigation of a mobile robot requires the recognition of the environment in terms of vision processing. To be guided in the given path, the robot should acquire the information about where the wall and corridor are located. Also unexpected obstacles should be detected as rapid as possible for the safe obstacle avoidance. In the paper, we assume that the mobile robot should be navigated in the flat surface. In terms of this assumption we simplify the correspondence problem by the free navigation surface and matching features in that coordinate system. Basically, the vision processing system adopts line segment of edge as the feature. The extracted line segments of edge out of both image are matched in the free nevigation surface. According to the matching result, each line segment is labeled by the attributes regarding obstacle and free surface and the 3D shape of obstacle is interpreted. This proposed vision processing method is verified in terms of various simulations and experimentation using real images.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2000년도 제15차 학술회의논문집
• /
• pp.555-555
• /
• 2000

This study is concentrated on the development of hight speed multi-object image processing algorithm, and based on these a1gorithm, vision control scheme is developed for the robot's position control in real time. Recently, the use of vision system is rapidly increasing in robot's position centre. To apply vision system in robot's position control, it is necessary to transform the physical coordinate of object into the image information acquired by CCD camera, which is called image processing. Thus, to control the robot's point position in real time, we have to know the center point of object in image plane. Particularly, in case of rigid body, the center points of multi-object must be calculated in a image plane at the same time. To solve these problems, the algorithm of multi-object for rigid body control is developed.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2014년도 추계학술발표대회
• /
• pp.1183-1186
• /
• 2014

Our research is different form traditional to have some large LED screen grouping together to constitute multi-vision technique. In this paper, we purpose a method of using k-path diffusion method to build connect between the devices and find an optimal data transmission path. In second half of this paper, through practical application, we using this technique transmitting data successfully and achieving a simple Multi-vision effect. This technique possess smart devices and Wifi P2P's features, these features improve system's dynamic and decentralized processing ability make our technique has high scalability.

• 반도체디스플레이기술학회지
• /
• 제21권1호
• /
• pp.75-78
• /
• 2022

Deep Neural Networks (DNN) has become an essential data processing architecture for the implementation of multiple computer vision tasks. Recently, DNN-based algorithms achieve much higher recognition accuracy than traditional algorithms based on shallow learning. However, training and inference DNNs require huge computational capabilities than daily usage purposes of computers. Moreover, with increased size and depth of DNNs, CPUs may be unsatisfactory since they use serial processing by default. GPUs are the solution that come up with greater speed compared to CPUs because of their Parallel Processing/Computation nature. In this paper, we analyze the inference time complexity of DNNs using well-known computer vision library, OpenCV. We measure and analyze inference time complexity for three cases, CPU, GPU-Float32, and GPU-Float16.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 2002년도 ICCAS
• /
• pp.41.5-41
• /
• 2002

This paper presents a feature extraction algorithm for vision-based micromanipulation. In order to guarantee of the accurate micromanipulation, most of micromanipulation systems use vision sensor. Vision data from an optical microscope or high magnification lens have vast information, however, characteristics of micro image such as emphasized contour, texture, and noise are make it difficult to apply macro image processing algorithms to micro image. Grasping points extraction is very important task in micromanipulation because inaccurate grasping points can cause breakdown of micro gripper or miss of micro objects. To solve those problems and extract grasping points for micromanipulation...

• 대한토목학회논문집
• /
• 제33권4호
• /
• pp.1683-1691
• /
• 2013

콘크리트 표면절삭 작업은 포장면의 노화 또는 파손으로 인한 보수작업과 그루빙(Grooving) 시공을 통한 포장면의 배수능력을 강화하거나 평탄성을 확보를 위하여 자주 적용되는 공법이다. 그러나 그 작업특성이 노동집약적이고 분진, 슬러지, 소음 등으로 인한 유해한 작업환경을 보유하고 있으며 장비를 다루는 기능공의 숙련도에 따라 생산성 및 절삭품질의 편차가 큰 경향이 있다. 따라서 장비 조종자가 각종 위험에 노출되지 않도록 하기 위한 원격조종 콘크리트 표면절삭 장비 개발이 필요하다. 원격 조종 환경에서는 조종자가 객관적인 절삭 품질을 확인함과 동시에 장비가 계획 경로에 따라 작업이 올바르게 수행되고 있는지를 확인할 수 있도록 하는 지원시스템이 필요케되는 바, 본 연구에서는 머신비전 시스템(Machine Vision System)과 GPS를 적용하여 네트워크 카메라로 촬영한 절삭면의 이미지를 디지털 영상처리(Image Processing)과정을 거쳐 객관적이며 품질관리 프로세스가 자동화된 시스템을 구축하였다. 또한 장비의 현재 위치와 방향, 속도, 계획된 경로와의 오차정보 그리고 작업의 진척도 등을 종합적으로 산출하여 워크 스테이션에 표시함과 동시에 머신 비전 시스템에 의한 작업 품질 정보와의 통합을 위한 프로그램을 개발하였으며, 현장 적용 테스트를 통해 본 기술을 검증하였다.