• International Journal of Computer Science & Network Security
• /
• v.22 no.2
• /
• pp.193-198
• /
• 2022

Face Detection is one of the most important aspects of image processing, it considers a time-consuming problem in real-time applications such as surveillance systems, face recognition systems, attendance system and many. At present, commodity hardware is getting more and more heterogeneity in terms of architectures such as GPU and MIC co-processors. Utilizing those co-processors along with the existing traditional CPUs gives the algorithm a better chance to make use of both architectures to achieve faster implementations. This paper presents a hybrid implementation of the face detection based on the local binary pattern (LBP) algorithm that is deployed on both traditional CPU and MIC co-processor to enhance the speed of the LBP algorithm. The experimental results show that the proposed implementation achieved improvement in speed by 3X when compared to a single architecture individually.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.19 no.3
• /
• pp.29-42
• /
• 2016

Remote sensing allows acquisition of information across a large area without contacting objects, and has thus been rapidly developed by application to different areas. Thus, with the development of remote sensing, satellites are able to rapidly advance in terms of their image resolution. As a result, satellites that use remote sensing have been applied to conduct research across many areas of the world. However, while research on remote sensing is being implemented across various areas, research on data processing is presently insufficient; that is, as satellite resources are further developed, data processing continues to lag behind. Accordingly, this paper discusses plans to maximize the performance of satellite image processing by utilizing the CUDA(Compute Unified Device Architecture) Library of NVIDIA, a parallel processing technique. The discussion in this paper proceeds as follows. First, standard KOMPSAT(Korea Multi-Purpose Satellite) images of various sizes are subdivided into five types. NDVI(Normalized Difference Vegetation Index) is implemented to the subdivided images. Next, ArcMap and the two techniques, each based on CPU or GPU, are used to implement NDVI. The histograms of each image are then compared after each implementation to analyze the different processing speeds when using CPU and GPU. The results indicate that both the CPU version and GPU version images are equal with the ArcMap images, and after the histogram comparison, the NDVI code was correctly implemented. In terms of the processing speed, GPU showed 5 times faster results than CPU. Accordingly, this research shows that a parallel processing technique using CUDA Library can enhance the data processing speed of satellites images, and that this data processing benefits from multiple advanced remote sensing techniques as compared to a simple pixel computation like NDVI.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.6
• /
• pp.591-596
• /
• 2019

In this paper, we implemented a deep learning operation structure with less influence of local PC performance. In general, the deep learning model has a large amount of computation and is heavily influenced by the performance of the processing PC. In this paper, we implemented deep learning operation using AWS and streaming server to reduce this limitation. First, deep learning operations were performed on AWS so that deep learning operation would work even if the performance of the local PC decreased. However, with AWS, the output is less real-time relative to the input when computed. Second, we use streaming server to increase the real-time of deep learning model. If the streaming server is not used, the real-time performance is poor because the images must be processed one by one or by stacking the images. We used the YOLO v3 model as a deep learning model for performance comparison experiments, and compared the performance of local PCs with instances of AWS and GTX1080, a high-performance GPU. The simulation results show that the test time per image is 0.023444 seconds when using the p3 instance of AWS, which is similar to the test time per image of 0.027099 seconds on a local PC with the high-performance GPU GTX1080.

• Journal of Positioning, Navigation, and Timing
• /
• v.7 no.4
• /
• pp.189-203
• /
• 2018

In this paper, a Software Defined Radio (SDR) architecture was designed and implemented for rapid prototyping of GNSS receiver. The proposed SDR can receive various GNSS and direct sequence spread spectrum (DSSS) signals without software modification by expanded input parameters containing information of the desired signal. Input parameters include code information, center frequency, message format, etc. To receive various signal by parameter controlling, a correlator, a data bit extractor and a receiver channel were designed considering the expanded input parameters. In navigation signal processing, pseudorange was measured based on Coordinated Universal Time (UTC) and appropriate navigation message decoder was selected by message format of input parameter so that receiver position can be calculated even if SDR is set up various GNSS combination. To validate the proposed SDR, the software was implemented using C++, CUDA C based on GPU and USRP. Experimentation has confirmed that changing the input parameters allows GPS, GLONASS, and BDS satellite signals to be received. The precision of the position from implemented SDR were measured below 5 m (Circular Error Probability; CEP) for all scenarios. This means that the implemented SDR operated normally. The implemented SDR will be used in a variety of fields by allowing prototyping of various GNSS signal only by changing input parameters.

• Journal of the Korea Computer Graphics Society
• /
• v.14 no.3
• /
• pp.1-9
• /
• 2008

This paper presents a novel optimization-based saliency-preserving method for converting color images to grayscale in a manner consistent with conventional approaches of black-and-white photographers. In black-and-white photography, a colored filter called a contrast filter has been commonly employed on a camera to lighten or darken selected colors. In addition, local exposure controls such as dodging and burning techniques are typically employed in the darkroom process to change the exposure of local areas within the print without affecting the overall exposure. Our method seeks a digital version of a conventional contrast filter to preserve visually-important image features. Furthermore, conventional burning and dodging techniques are addressed, together with image similarity weights, to give edge-aware local exposure control over the image space. Our method can be efficiently optimized on GPU. According to the experiments, CUDA implementation enables 1 megapixel color images to be converted to grayscale at interactive frames rates.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.6
• /
• pp.1143-1148
• /
• 2017

Currently, Artificial Intelligence and Deep Learning are rising as hot social issues, and these technologies are applied to various fields. A good method among the various algorithms in Artificial Intelligence is Convolutional Neural Networks. Convolutional Neural Network is a form that adds Convolution Layers to Multi Layer Neural Network. If you use Convolutional Neural Networks for small amount of data, or if the structure of layers is not complicated, you don't have to pay attention to speed. But the learning should take long time when the size of the learning data is large and the structure of layers is complicated. In these cases, GPU-based parallel processing is frequently needed. In this paper, we developed Convolutional Neural Networks using CUDA, and show that its learning is faster and more efficient than learning using some other frameworks or programs.

• Journal of the Korea Computer Graphics Society
• /
• v.23 no.3
• /
• pp.19-29
• /
• 2017

When a fixed-sized memory allocation method is used for sparse volume data, a considerable memory space is in general wasted, which becomes more serious for a large volume of high resolution. In this paper, in order to reduce such unnecessary memory consumption, we propose a volume representation method to store mostly voxels that represent valid information rather than all voxels in a fixed volume space. Then our method is compared with the conventional static memory allocation method, an octree-based representation, and a voxel hashing method in terms of memory usage and computation speed. In particular, we compare the proposed method and the voxel hashing method with respect to implementation of the GPU-based Marching Cubes algorithm.

• The Journal of the Korea Contents Association
• /
• v.14 no.10
• /
• pp.1-9
• /
• 2014

Generally, integral image used for autostereoscopic 3d display is generated for flat lens array, but flat lens array cannot provide a wide range of view for generated integral image because of narrow range of view. To make up for this flat lens array's weak point, curved lens array has been proposed, and due to technical and cost problem, approximate surface lens array composed of several flat lens array is used instead of ideal curved lens array. In this paper, we constructed an approximate surface lens array arranged for $20{\times}8$ square flat lens in 100mm radius sphere, and we could get about twice angle of view compared to flat lens array. Specially, unlike existing researches which manually generate integral image, we propose an OpenCL GPU parallel process algorithm for generating real-time integral image. As a result, we could get 12-20 frame/sec speed about various 3D volume data from $15{\times}15$ approximate surface lens array.

• Journal of Broadcast Engineering
• /
• v.25 no.5
• /
• pp.655-664
• /
• 2020

Free viewpoint image synthesis technology is one of the important technologies in the MPEG-I (Immersive) standard. RVS (Reference View Synthesizer) developed by MPEG-I and in use in MPEG group is a DIBR (Depth Information-Based Rendering) program that generates an image at a virtual (intermediate) viewpoint from multiple viewpoints' inputs. RVS uses the mesh surface method based on computer graphics, and outperforms the pixel-based ones by 2.5dB or more compared to the previous pixel method. Even though its OpenGL version provides 10 times speed up over the non OpenGL based one, it still shows a non-real-time processing speed, i.e., 0.75 fps on the two 2k resolution input images. In this paper, we analyze the internal of RVS implementation and modify its structure, achieving 34 times speed up, therefore, real-time performance (22-26 fps), through the 3 key improvements: 1) the reuse of OpenGL buffers and texture objects 2) the parallelization of file I/O and OpenGL execution 3) the parallelization of GPU shader program and buffer transfer.

• Journal of IKEEE
• /
• v.24 no.1
• /
• pp.225-231
• /
• 2020

In this paper, we propose a method to accelerate BWN based on FPGA with limited resources for embedded system. Because of the limited number of logic elements available, a single computing unit capable of handling Conv-layer, FC-layer of various sizes must be designed and reused. Also, if the input feature map can not be parallel processed at one time, the output must be calculated by reading the inputs several times. Since the number of available BRAM modules is limited, the number of data bits in the BWN accelerator must be minimized. The image classification processing time of the BWN accelerator is superior when compared with a embedded CPU and is faster than a desktop PC and 50% slower than a GPU system. Since the BWN accelerator uses a slow clock of 50MHz, it can be seen that the BWN accelerator is advantageous in performance versus power.