• IEMEK Journal of Embedded Systems and Applications
• /
• v.15 no.2
• /
• pp.51-60
• /
• 2020

Optical character recognition (OCR) has been studied for decades because it is very useful in a variety of places. Nowadays, OCR's performance has improved significantly due to outstanding deep learning technology. Thus, there is an increasing demand for commercial-grade but affordable OCR systems. We have developed a low-cost, high-performance OCR system for the industry with the cheapest embedded developer kit that supports GPU acceleration. To achieve high accuracy for industrial use on limited computing resources, we chose a state-of-the-art text recognition algorithm that uses an end-to-end deep learning network as a baseline model. The model was then improved by replacing the feature extraction network with the best one suited to our conditions. Among the various candidate networks, EfficientNet-B3 has shown the best performance: excellent recognition accuracy with relatively low memory consumption. Besides, we have optimized the model written in TensorFlow's Python API using TensorFlow-TensorRT integration and TensorFlow's C++ API, respectively.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.8
• /
• pp.253-263
• /
• 2013

Frequency Scanning Interferometry(FSI) system, one of the most promising optical surface measurement techniques, generally results in superior optical performance comparing with other 3-dimensional measuring methods as its hardware structure is fixed in operation and only the light frequency is scanned in a specific spectral band without vertical scanning of the target surface or the objective lens. FSI system collects a set of images of interference fringe by changing the frequency of light source. After that, it transforms intensity data of acquired image into frequency information, and calculates the height profile of target objects with the help of frequency analysis based on Fast Fourier Transform(FFT). However, it still suffers from optical noise on target surfaces and relatively long processing time due to the number of images acquired in frequency scanning phase. 1) a Polarization-based Frequency Scanning Interferometry(PFSI) is proposed for optical noise robustness. It consists of tunable laser for light source, ${\lambda}/4$ plate in front of reference mirror, ${\lambda}/4$ plate in front of target object, polarizing beam splitter, polarizer in front of image sensor, polarizer in front of the fiber coupled light source, ${\lambda}/2$ plate between PBS and polarizer of the light source. Using the proposed system, we can solve the problem of fringe image with low contrast by using polarization technique. Also, we can control light distribution of object beam and reference beam. 2) the signal processing acceleration method is proposed for PFSI, based on parallel processing architecture, which consists of parallel processing hardware and software such as Graphic Processing Unit(GPU) and Compute Unified Device Architecture(CUDA). As a result, the processing time reaches into tact time level of real-time processing. Finally, the proposed system is evaluated in terms of accuracy and processing speed through a series of experiment and the obtained results show the effectiveness of the proposed system and method.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 2017.04a
• /
• pp.160-160
• /
• 2017

자동화 기술을 통한 한국형 스마트팜의 발전이 비약적으로 이루어지고 있는 가운데 무인화를 위한 지능적인 스마트 시설환경 관찰 및 분석에 대한 요구가 점점 증가 하고 있다. 스마트 시설환경에서 취득 가능한 시계열 데이터는 온도, 습도, 조도, CO2, 토양 수분, 환기량 등 다양하다. 시스템의 경계가 명확함에도 해당 속성의 특성상 타임도메인과 공간도메인 상에서 정확한 추정 또는 예측이 난해하다. 시설 환경에 접목이 증가하고 있는 지능형 관리 기술 구현을 위해선 시계열 공간 데이터에 대한 신속하고 정확한 정량화 기술이 필수적이라 할 수 있다. 이러한 기술적인 요구사항을 해결하고자 시도되는 다양한 방법 중에서 공간 분해능 향상을 위한 다지점 계측 메트릭스를 실험적으로 구성하였다. $50m{\times}100m$의 단면적인 연동 딸기 온실을 대상으로 $3{\times}3{\times}3$의 3차원 환경 인자 계측 매트릭스를 설치하였다. 1 Hz의 주기로 4가지 환경인자(온도, 습도, 조도, CO2)를 계측하였으며, 계측 하는 시점과 동시에 병렬적으로 공간통계법을 이용하여 미지의 지점에 대한 환경 인자들을 실시간으로 추정하였다. 선행적으로 50 cm 공간 분해능에 대응하기 위하여 Kriging interpolation법을 횡단면에 대하여 분석한 후 다시 종단면에 대하여 분석하였다. 3 Ghz에 해당하는 연산 능력을 보유한 컴퓨터에서 1초 동안 획득한 데이터에 대한 분석을 마치는데 소요되는 시간이 15초 내외로 나타났다. 이는 해당 알고리즘의 매우 높은 시간 복잡도(Order of $O=O^3$)에 기인하는 것으로 다양한 시설 환경의 관리 방법론에 적절히 대응하기에 한계가 있다 할 수 있다. 실시간으로 시간 복잡도가 높은 연산을 수행하기 위한 기술적인 과제를 해결하고자, 근래에 관심이 증가하고 있는 NVIDIA 사에서 제공하는 CUDA 엔진과 Apple사의 제안을 시작으로 하여 공개 소프트웨어 개발 컨소시엄인 크로노스 그룹에서 제공하는 OpenCL 엔진을 비교 분석하였다. CUDA 엔진은 GPU(Graphics Processing Unit)에서 정보 분석 프로그램의 연산 집약적인 부분만을 담당하여 신속한 결과를 산출할 수 있는 라이브러리이며 해당 하드웨어를 구비하였을 때 사용이 가능하다. 반면, OpenCL은 CUDA 엔진이 특정 하드웨어에서 구동이 되는 한계를 극복하고자 하드웨어에 비의존적인 라이브러리를 제공하는 것이 다르며 클러스터링 기술과 연계를 통해 낮은 하드웨어 성능으로 인한 단점을 극복하고자 하였다. 본 연구에서는 CUDA 8.0(https://developer.nvidia.com/cuda-downloads)버전과 Pascal Titan X(NVIDIA, CA, USA)를 사용한 방법과 OpenCL 1.2(https://www.khronos.org/opencl/)버전과 Samsung Exynos5422 칩을 장착한 ODROID-XU4(Hardkernel, AnYang, Korea)를 사용한 방법을 비교 분석하였다. 50 cm의 공간 분해능에 대응하기 위한 4차원 행렬($100{\times}200{\times}5{\times}4$)에 대하여 정수 지수화를 위한 Quantization을 거쳐 CUDA 엔진과 OpenCL 엔진을 적용한 비교한 결과, CUDA 엔진은 1초 내외, OpenCL 엔진의 경우 5초 내외의 연산 속도를 보였다. CUDA 엔진의 경우 비용측면에서 약 10배, 전력 소모 측면에서 20배 이상 소요되었다. 따라서 우선적으로 OpenCL 엔진 기반 하드웨어 가속 기술 최적화 연구를 통해 스마트 시설환경 실시간 시뮬레이션 기술 도입을 위한 기술적 과제를 풀어갈 것이다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2015.05a
• /
• pp.270-273
• /
• 2015

In shipbuilding and marine industry, engineering and design software solutions have upgraded from the original 2D schematic data based CAD system to a modern 3D drawing-based system. Due to the fact that the massive amount of data usage in real time and data volumes of various engineering models including graphic data have increased, several problems such as lack of server resources and improper handling of 3D drawings have been raised. Besides, increasing the number of session connections per server can cause deterioration of server performance. Recently, increasing the yard's sophisticated design capabilities highlighted the need to develop engineering and design system which would not only overcome the network performance issues, but would provide efficient collaborative design environment. This paper presents an overview of the framework for collaborative engineering design system based on the virtual application (Citrix XenApp 6.5)and acceleration hardware technology of 3D graphics (NVIDIA GRID K2 solution).

• The Journal of the Korea Contents Association
• /
• v.11 no.3
• /
• pp.9-17
• /
• 2011

Recently, with the advent of stereoscopic 3D TV, the activation of 3D stereoscopic content is expected. Research on 3D auto stereoscopic display has been carried out to relieve discomfort of 3D stereoscopic display. In this research, it is necessary to generate the elemental image from a lens array. As the number of lens in a lens array is increased, it takes a lot of time to generate the elemental image, and it will take more time for a large volume data. In order to improve the problem, in this paper, we propose a method to generate the elemental image by using OpenCL based on CUDA. We perform our proposed method on PC environment with one of Tesla C1060, Geforce 9800GT and Quadro FX 3800 graphics cards. Experimental results show that the proposed method can obtain almost 20 times better performance than recent research result[11].

• Journal of Broadcast Engineering
• /
• v.26 no.6
• /
• pp.778-789
• /
• 2021

Recently, various lightweight methods for using Convolutional Neural Network(CNN) models in mobile devices have emerged. Weight quantization, which lowers bit precision of weights, is a lightweight method that enables a model to be used through integer calculation in a mobile environment where GPU acceleration is unable. Weight quantization has already been used in various models as a lightweight method to reduce computational complexity and model size with a small loss of accuracy. Considering the size of memory and computing speed as well as the storage size of the device and the limited network environment, this paper proposes a method of compressing integer weights after quantization using a video codec as a method. To verify the performance of the proposed method, experiments were conducted on VGG16, Resnet50, and Resnet18 models trained with ImageNet and Places365 datasets. As a result, loss of accuracy less than 2% and high compression efficiency were achieved in various models. In addition, as a result of comparison with similar compression methods, it was verified that the compression efficiency was more than doubled.

• Journal of the Korea Computer Graphics Society
• /
• v.28 no.3
• /
• pp.113-123
• /
• 2022

Ray tracing-based rendering creates by far more realistic images than the traditional rasterization-based rendering. However, it is still burdensome when implemented for a Head-Mounted Display (HMD) system that demands a wide field of view and a high display refresh rate. Furthermore, for presenting high-quality images on the HMD screen, a sufficient number of ray sampling should be carried out per pixel to alleviate visually annoying spatial and temporal aliases. In this paper, we extend the recent selective foveated ray tracing technique by Kim et al. [1], and propose an improved real-time rendering technique that realizes the rendering effect of the classic Whitted-style ray tracing on the HMD system. In particular, by combining the ray tracing hardware-based acceleration technique and time-constrained rendering scheme, we show that fast HMD ray tracing is possible that is well suited to human visual systems.

• Journal of the Korean Society of Radiology
• /
• v.17 no.5
• /
• pp.633-640
• /
• 2023

With the recent development of static and dynamic modulated brachytherapy methods in brachytherapy, which use radiation shielding to modulate the dose distribution to deliver the dose, the amount of parameters and data required for dose calculation in inverse treatment planning and treatment plan optimization algorithms suitable for new directional beam intensity modulated brachytherapy is increasing. Although intensity-modulated brachytherapy enables accurate dose delivery of radiation, the increased amount of parameters and data increases the elapsed time required for dose calculation. In this study, a GPU-based CUDA-accelerated dose calculation algorithm was constructed to reduce the increase in dose calculation elapsed time. The acceleration of the calculation process was achieved by parallelizing the calculation of the system matrix of the volume of interest and the dose calculation. The developed algorithms were all performed in the same computing environment with an Intel (3.7 GHz, 6-core) CPU and a single NVIDIA GTX 1080ti graphics card, and the dose calculation time was evaluated by measuring only the dose calculation time, excluding the additional time required for loading data from disk and preprocessing operations. The results showed that the accelerated algorithm reduced the dose calculation time by about 30 times compared to the CPU-only calculation. The accelerated dose calculation algorithm can be expected to speed up treatment planning when new treatment plans need to be created to account for daily variations in applicator movement, such as in adaptive radiotherapy, or when dose calculation needs to account for changing parameters, such as in dynamically modulated brachytherapy.