• Journal of Korea Multimedia Society
• /
• v.16 no.10
• /
• pp.1189-1195
• /
• 2013

In this paper, we are going to describe the performance characteristic of very high resolution image processing with WebCL on the web environment. In order to evaluate the variance of the execution time by WebCL, we modified the Pixastic library, one of the most representative image processing libraries written in JavaScript, by using WebCL. We achieved a speedup of up to 4.2 times and 2.8 times on average against the original one for the image of 8K Ultra HD with the WebCL-based library.

• Korean Journal of Remote Sensing
• /
• v.36 no.5_3
• /
• pp.1027-1036
• /
• 2020

In this study, a cloud-based processing method using Agisoft Metashape, a commercial software, and Amazon web service, a cloud computing service, is introduced and evaluated to quickly generate high-precision 3D realistic data from large volume UAV images acquired in disaster sites. Compared with on-premises method using a local computer and cloud services provided by Agisoft and Pix4D, the processes of aerial triangulation, 3D point cloud and DSM generation, mesh and texture generation, ortho-mosaic image production recorded similar time duration. The cloud method required uploading and downloading time for large volume data, but it showed a clear advantage that in situ processing was practically possible. In both the on-premises and cloud methods, there is a difference in processing time depending on the performance of the CPU and GPU, but notso much asin a performance benchmark. However, it wasfound that a laptop computer equipped with a low-performance GPU takes too much time to apply to in situ processing.

• Annual Conference of KIPS
• /
• 2019.10a
• /
• pp.19-20
• /
• 2019

서버리스 프레임워크(Serverless Framework)는 마이크로서비스 아키텍처의 이론을 클라우드와 컨테이너를 기반으로 구현한 것으로 아마존의 AWS(Amazon Web Service)와 같은 퍼블릭 클라우드 플랫폼이 서비스됨에 따라 활용도 높아지고 있다. 하지만 현재까지의 플랫폼들은 GPU 와 같은 하드웨어의 의존성을 가진 인공지능 모델의 서비스에는 지원이 부족하다. 이에 본 논문에서는 컨테이너 기반의 오픈소스 서버리스 플랫폼을 대상으로 엔비디어-도커와 k8s-device-plugin 을 적용하여 GPU 활용이 가능한 서버리스 플랫폼을 구현하였다. 또한 인공지능 모델이 컨테이너에서 구동될 때 반복되는 가중치 로드를 줄이기 위한 구조를 제안한다. 본 논문에서 구현된 서버리스 플랫폼은 객체 검출 모델인 SSD(Single Shot Multibox Detector) 모델을 이용하여 성능 비교 실험을 진행하였으며, 그 결과 인공지능 모델이 적용된 서버리스 플랫폼의 함수 응답 시간이 개선되었음을 확인하였다.

• Journal of Korea Game Society
• /
• v.11 no.1
• /
• pp.147-157
• /
• 2011

Webcam doesn't perform memory-alignment in order to reduce the transmission time of image data. Memory-unaligned image data is unsuitable for the processing on GPU. Accordingly, we convert it to available color format for optimization in high speed image processing. In this paper, we propose a technique that accelerates webcam's color format conversion by using NVDIA CUDA. We propose an optimization which is about memory accesses and thread composition, also evaluate memory and computing performance for verifying a hypothesis which is the performance of the proposed architecture and optimizing degree on low-performance GPU. Following the optimization technique, we show performance improvements over maximum 68 percent.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.9
• /
• pp.1279-1292
• /
• 2022

This study proposes a non-contact input method based on face recognition and Pyautogui mouse control as a system that can help users who have difficulty using input devices such as conventional mouse due to physical discomfort. This study includes features that help web surfing more conveniently, especially screen zoom, scroll function, and also solves the problem of eye fatigue, which has been suggested as a limitation in existing non-contact input systems. In addition, various set values can be adjusted in consideration of individual physical differences and Internet usage habits. Furthermore, no high-performance CPU or GPU environment is required, and no separate tracker devices or high-performance cameras are required. Through these studies, we intended to contribute to the realization of barrier-free access by increasing the web accessibility of the disabled and the elderly who find it difficult to use web content.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.7
• /
• pp.31-45
• /
• 2019

The core technologies for intelligent services today are deep learning, that is neural networks, and parallel and distributed processing technologies such as GPU parallel computing and big data. However, for intelligent services and knowledge sharing services through globally shared ontologies in the future, there is a technology that is better than the neural networks for representing and reasoning knowledge. It is a knowledge representation of IF-THEN in RIF or SWRL, which is the standard rule language of the Semantic Web, and can be inferred efficiently using the rete algorithm. However, when the number of rules processed by the rete algorithm running on a single computer is 100,000, its performance becomes very poor with several tens of minutes, and there is an obvious limitation. Therefore, in this paper, we analyze the past and current studies on parallel and distributed processing of rete algorithm, and examine what aspects should be considered to implement an efficient rete algorithm.

• Journal of Intelligence and Information Systems
• /
• v.24 no.2
• /
• pp.195-220
• /
• 2018

In this study, we conducted an empirical analysis of the factors that affect the change of Bitcoin Closing Price. Previous studies have focused on the security of the block chain system, the economic ripple effects caused by the cryptocurrency, legal implications and the acceptance to consumer about cryptocurrency. In various area, cryptocurrency was studied and many researcher and people including government, regardless of country, try to utilize cryptocurrency and applicate to its technology. Despite of rapid and dramatic change of cryptocurrencies' price and growth of its effects, empirical study of the factors affecting the price change of cryptocurrency was lack. There were only a few limited studies, business reports and short working paper. Therefore, it is necessary to determine what factors effect on the change of closing Bitcoin price. For analysis, hypotheses were constructed from three dimensions of consumer, industry, and macroeconomics for analysis, and time series data were collected for variables of each dimension. Consumer variables consist of search traffic of Bitcoin, search traffic of bitcoin ban, search traffic of ransomware and search traffic of war. Industry variables were composed GPU vendors' stock price and memory vendors' stock price. Macro-economy variables were contemplated such as U.S. dollar index futures, FOMC policy interest rates, WTI crude oil price. Using above variables, we did times series regression analysis to find relationship between those variables and change of Bitcoin Closing Price. Before the regression analysis to confirm the relationship between change of Bitcoin Closing Price and the other variables, we performed the Unit-root test to verifying the stationary of time series data to avoid spurious regression. Then, using a stationary data, we did the regression analysis. As a result of the analysis, we found that the change of Bitcoin Closing Price has negative effects with search traffic of 'Bitcoin Ban' and US dollar index futures, while change of GPU vendors' stock price and change of WTI crude oil price showed positive effects. In case of 'Bitcoin Ban', it is directly determining the maintenance or abolition of Bitcoin trade, that's why consumer reacted sensitively and effected on change of Bitcoin Closing Price. GPU is raw material of Bitcoin mining. Generally, increasing of companies' stock price means the growth of the sales of those companies' products and services. GPU's demands increases are indirectly reflected to the GPU vendors' stock price. Making an interpretation, a rise in prices of GPU has put a crimp on the mining of Bitcoin. Consequently, GPU vendors' stock price effects on change of Bitcoin Closing Price. And we confirmed U.S. dollar index futures moved in the opposite direction with change of Bitcoin Closing Price. It moved like Gold. Gold was considered as a safe asset to consumers and it means consumer think that Bitcoin is a safe asset. On the other hand, WTI oil price went Bitcoin Closing Price's way. It implies that Bitcoin are regarded to investment asset like raw materials market's product. The variables that were not significant in the analysis were search traffic of bitcoin, search traffic of ransomware, search traffic of war, memory vendor's stock price, FOMC policy interest rates. In search traffic of bitcoin, we judged that interest in Bitcoin did not lead to purchase of Bitcoin. It means search traffic of Bitcoin didn't reflect all of Bitcoin's demand. So, it implies there are some factors that regulate and mediate the Bitcoin purchase. In search traffic of ransomware, it is hard to say concern of ransomware determined the whole Bitcoin demand. Because only a few people damaged by ransomware and the percentage of hackers requiring Bitcoins was low. Also, its information security problem is events not continuous issues. Search traffic of war was not significant. Like stock market, generally it has negative in relation to war, but exceptional case like Gulf war, it moves stakeholders' profits and environment. We think that this is the same case. In memory vendor stock price, this is because memory vendors' flagship products were not VRAM which is essential for Bitcoin supply. In FOMC policy interest rates, when the interest rate is low, the surplus capital is invested in securities such as stocks. But Bitcoin' price fluctuation was large so it is not recognized as an attractive commodity to the consumers. In addition, unlike the stock market, Bitcoin doesn't have any safety policy such as Circuit breakers and Sidecar. Through this study, we verified what factors effect on change of Bitcoin Closing Price, and interpreted why such change happened. In addition, establishing the characteristics of Bitcoin as a safe asset and investment asset, we provide a guide how consumer, financial institution and government organization approach to the cryptocurrency. Moreover, corroborating the factors affecting change of Bitcoin Closing Price, researcher will get some clue and qualification which factors have to be considered in hereafter cryptocurrency study.

• Journal of Digital Convergence
• /
• v.16 no.10
• /
• pp.205-212
• /
• 2018

With the development of high performance CPU / GPU, artificial intelligence algorithms such as deep neural networks, and a large amount of data, machine learning has been extended to various applications. In particular, a large amount of data collected from the Internet of Things, social network services, web pages, and public data is accelerating the use of machine learning. Learning data sets for machine learning exist in various formats according to application fields and data types, and thus it is difficult to effectively process data and apply them to machine learning. Therefore, this paper studied a method for building a learning data set for machine learning in accordance with standardized procedures. This paper first analyzes the requirement of learning data set according to problem types and data types. Based on the analysis, this paper presents the reference model to build learning data set for machine learning applications. This paper presents the target standardization organization and a standard development strategy for building learning data set.

• Journal of Intelligence and Information Systems
• /
• v.26 no.2
• /
• pp.1-25
• /
• 2020

In this paper, we suggest an application system architecture which provides accurate, fast and efficient automatic gasometer reading function. The system captures gasometer image using mobile device camera, transmits the image to a cloud server on top of private LTE network, and analyzes the image to extract character information of device ID and gas usage amount by selective optical character recognition based on deep learning technology. In general, there are many types of character in an image and optical character recognition technology extracts all character information in an image. But some applications need to ignore non-of-interest types of character and only have to focus on some specific types of characters. For an example of the application, automatic gasometer reading system only need to extract device ID and gas usage amount character information from gasometer images to send bill to users. Non-of-interest character strings, such as device type, manufacturer, manufacturing date, specification and etc., are not valuable information to the application. Thus, the application have to analyze point of interest region and specific types of characters to extract valuable information only. We adopted CNN (Convolutional Neural Network) based object detection and CRNN (Convolutional Recurrent Neural Network) technology for selective optical character recognition which only analyze point of interest region for selective character information extraction. We build up 3 neural networks for the application system. The first is a convolutional neural network which detects point of interest region of gas usage amount and device ID information character strings, the second is another convolutional neural network which transforms spatial information of point of interest region to spatial sequential feature vectors, and the third is bi-directional long short term memory network which converts spatial sequential information to character strings using time-series analysis mapping from feature vectors to character strings. In this research, point of interest character strings are device ID and gas usage amount. Device ID consists of 12 arabic character strings and gas usage amount consists of 4 ~ 5 arabic character strings. All system components are implemented in Amazon Web Service Cloud with Intel Zeon E5-2686 v4 CPU and NVidia TESLA V100 GPU. The system architecture adopts master-lave processing structure for efficient and fast parallel processing coping with about 700,000 requests per day. Mobile device captures gasometer image and transmits to master process in AWS cloud. Master process runs on Intel Zeon CPU and pushes reading request from mobile device to an input queue with FIFO (First In First Out) structure. Slave process consists of 3 types of deep neural networks which conduct character recognition process and runs on NVidia GPU module. Slave process is always polling the input queue to get recognition request. If there are some requests from master process in the input queue, slave process converts the image in the input queue to device ID character string, gas usage amount character string and position information of the strings, returns the information to output queue, and switch to idle mode to poll the input queue. Master process gets final information form the output queue and delivers the information to the mobile device. We used total 27,120 gasometer images for training, validation and testing of 3 types of deep neural network. 22,985 images were used for training and validation, 4,135 images were used for testing. We randomly splitted 22,985 images with 8:2 ratio for training and validation respectively for each training epoch. 4,135 test image were categorized into 5 types (Normal, noise, reflex, scale and slant). Normal data is clean image data, noise means image with noise signal, relfex means image with light reflection in gasometer region, scale means images with small object size due to long-distance capturing and slant means images which is not horizontally flat. Final character string recognition accuracies for device ID and gas usage amount of normal data are 0.960 and 0.864 respectively.