• Journal of Broadcast Engineering
• /
• v.27 no.3
• /
• pp.341-350
• /
• 2022

In this paper, we propose a deep learning based on-device augmented reality (AR) system in which multiple input images are used to implement the correct occlusion in a real environment. The proposed system is composed of three technical steps; camera pose estimation, depth estimation, and object augmentation. Each step employs various mobile frameworks to optimize the processing on the on-device environment. Firstly, in the camera pose estimation stage, the massive computation involved in feature extraction is parallelized using OpenCL which is the GPU parallelization framework. Next, in depth estimation, monocular and multiple image-based depth image inference is accelerated using the mobile deep learning framework, i.e. TensorFlow Lite. Finally, object augmentation and occlusion handling are performed on the OpenGL ES mobile graphics framework. The proposed augmented reality system is implemented as an application in the Android environment. We evaluate the performance of the proposed system in terms of augmentation accuracy and the processing time in the mobile as well as PC environments.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.21 no.8
• /
• pp.1024-1033
• /
• 1997

The past several years have witnessed an ever-increasing acceptance and adoption of parallel processing, both for high performance scientific computing as well as for more general purpose applications. Furthermore with increasing needs to perform the complex flow calculations in an efficient manner, the use of the message passing model on distributed networks has emerged as an important alternative to the expensive supercomputers. This work attempts to provide a generic framework to enable the parallelization of all CFD-related works using the master-slave model. This framework consists of (1) input geometry, (2) domain decomposition, (3) grid generation, (4) flow computations, (5) flow visualization, and (6) output display as the sequential components, but performs computations for (2) to (5) in parallel on the workstation clustering. The flow computations are parallized by having multiple copies of the flow-code to solve a PDE on different spatial regions on different processors, while their flow data are exchanged across the region boundaries, and the solution is time-stepped. The Parallel Virtual Machine (PVM) is used for distributed communication in this work.

• Journal of the Institute of Convergence Signal Processing
• /
• v.6 no.4
• /
• pp.182-190
• /
• 2005

In the paper, we propose an intelligent automatic parking assist system. To realize an automatic parking, first, the prospective parking position and the location of a vehicle should be recognized. Second, the system should compute a path which introduces the parking position precisely with avoiding any obstacles. Third, the handle should be controlled so that the vehicle moves through the path. To calculate the location of the vehicle and its surroundings, the system applies the camera image method to transforming input images to the plane map. It also uses the inertial navigation method which recognizes the position and the direction of a moving vehicle by using a kinematic model of the vehicle. To generate a path of the vehicle, the simple path method and the Bezier spline method are tested. The divided arc method which generates multiple paths is also tested. We apply a method which makes the system choose the best path with multiple objective functions. We introduce the virtual road method, as a solution for the problem of mechanical time delay, to have the vehicle followed the designated path.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.19 no.9
• /
• pp.2121-2128
• /
• 2015

In this paper presents a method of extracting reliable and regions of interest (ROI) in barcode for the purpose of factory automation. backgrounds are separated based on directional components and the characteristics of detected patterns. post-processing is performed on candidate images with analysis of problems caused by blur, rotation and areas of high similarity. In addition, the resizing factor is used to achieve faster calculations through image resizing. The input images contained multiple product or barcode for application to diverse automation environments; a high extraction success rate is accomplished despite the maximum shooting distance of 80 cm. Simulations involving images with various shooting distances gave an ROI detection rate of 100% and a post-processing success rate of 99.3%.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.49 no.4
• /
• pp.49-57
• /
• 2012

This paper presents a fast multiple-image-based deblurring method that decreases the computation loads in the image deblurring, enhancing the sharpness of the textures or edges of the restored images. First, two blurred images with some blurring artifacts and one noisy image including severe noises are consecutively captured under a relatively long and short exposures, respectively. To improve the processing speeds, the captured multiple images are downsampled at the ratio of two, and then a way of estimating the point spread function(PSF) based on the image or edge patches extracted from the whole images, is introduced. The method enables to effectively reduce the computation time taken in the PSF prediction. Next, the texture-enhanced image deblurring method of supplementing the ability of the texture representation degraded by the downsampling of the input images, is developed and then applied. Finally, to get the same image size as the original input images, an upsampling method of utilizing the sharp edges of the captured noisy image is applied. By using the proposed method, the processing times taken in the image deblurring, which is the main obstacle of its application to the digital cameras, can be shortened, while recovering the fine details of the textures or edge components.

• The Journal of the Acoustical Society of Korea
• /
• v.23 no.6
• /
• pp.488-496
• /
• 2004

This paper describes a music information retrieval system which uses humming as the key for retrieval Humming is an easy way for the user to input a melody. However, there are several problems with humming that degrade the retrieval of information. One problem is a human factor. Sometimes people do not sing accurately, especially if they are inexperienced or unaccompanied. Another problem arises from signal processing. Therefore, a music information retrieval method should be sufficiently robust to surmount various humming errors and signal processing problems. A retrieval system has to extract pitch from the user's humming. However pitch extraction is not perfect. It often captures half or double pitches. even if the extraction algorithms take the continuity of the pitch into account. Considering these problems. we propose a system that takes multiple pitch candidates into account. In addition to the frequencies of the pitch candidates. the confidence measures obtained from their powers are taken into consideration as well. We also propose the use of an algorithm with three dimensions that is an extension of the conventional DP algorithm, so that multiple pitch candidates can be treated. Moreover in the proposed algorithm. DP paths are changed dynamically to take deltaPitches and IOIratios of input and reference notes into account in order to treat notes being split or unified. We carried out an evaluation experiment to compare the proposed system with a conventional system. From the experiment. the proposed method gave better retrieval performance than the conventional system.

• Design & Manufacturing
• /
• v.16 no.3
• /
• pp.50-57
• /
• 2022

The qualities of the products produced by injection molding are strongly influenced by the process variables of the injection molding machine set by the engineer. It is very difficult to predict the qualities of the injection molded product considering the stochastic nature of the manufacturing process, since the processing conditions have a complex impact on the quality of the injection molded product. It is recognized that the artificial neural network(ANN) is capable of mapping the intricate relationship between the input and output variables very accurately, therefore, many studies are being conducted to predict the relationship between the results of the product and the process variables using ANN. However in the condition of a small number of data sets, the predicting performance and robustness of the ANN model could be reduced due to too many input variables. In the present study, the ANN model that predicts the length of the injection molded product for multiple combinations of process variables was developed. And the accuracy of each ANN model was compared for 8 process variables and 4 important process inputs that were determined by the feature selection. Based on the comparison, it was verified that the performance of the ANN model increased when only 4 important variables were applied.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.3
• /
• pp.678-700
• /
• 2023

Diagnosing breast diseases using ultrasound (US) images remains challenging because it is time-consuming and requires expert radiologist knowledge. As a result, the diagnostic performance is significantly biased. To assist radiologists in this process, computer-aided diagnosis (CAD) systems have been developed and used in practice. This type of system is used not only to assist radiologists in examining breast ultrasound images (BUS) but also to ensure the effectiveness of the diagnostic process. In this study, we propose a new approach for breast lesion localization and segmentation using a multi-scale pyramid of the ultrasound image of a breast organ and a convolutional semantic segmentation network. Unlike previous studies that used only a deep detection/segmentation neural network on a single breast ultrasound image, we propose to use multiple images generated from an input image at different scales for the localization and segmentation process. By combining the localization/segmentation results obtained from the input image at different scales, the system performance was enhanced compared with that of the previous studies. The experimental results with two public datasets confirmed the effectiveness of the proposed approach by producing superior localization/segmentation results compared with those obtained in previous studies.

• KIPS Transactions on Software and Data Engineering
• /
• v.7 no.10
• /
• pp.397-404
• /
• 2018

Various applications are widely developed for smartphones to meet customer's needs. In many companies, messenger's typed interactive systems have been studied for business marketing, advertising and promotion to provide useful services for the customers. Such interactive systems are usually called as "Chatbot". In Chatbot, duplicated responses from Chatbot could occur frequently, and these make one lose interest. In this paper, we define a case that the response of Chatbot is duplicated according to the user's input, and propose a method to reduce duplicated responses of Chatbot. In the proposed method, we try to reduce duplication responses through a new duplication avoidance algorithm by building multiple tables in a database and by making combinations of user's input and its response in each table. In our experiments, the proposed method shows that duplicated responses are reduced by an average of 70%, compared with the existing method.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.777-784
• /
• 2003

In this Paper. we propose a recognition method of machine printed characters based on the combination of recognition units using multiple neural networks. In our recognition method, the input character is classified into one of 7 character types among which the first 6 types are for Hangul character and the last type is for non-Hangul characters. Hangul characters are recognized by several MLP (multilayer perceptron) neural networks through two stages. In the first stage, we divide Hangul character image into two or three recognition units (HRU : Hangul recognition unit) according to the combination fashion of graphemes. Each recognition unit composed of one or two graphemes is recognized by an MLP neural network with an input feature vector of pixel direction angles. In the second stage, the recognition aspect features of the HRU MLP recognizers in the first stage are extracted and forwarded to a subsequent MLP by which final recognition result is obtained. For the recognition of non-Hangul characters, a single MLP is employed. The recognition experiments had been performed on the character image database collected from 50,000 real letter envelope images. The experimental results have demonstrated the superiority of the proposed method.