• The Journal of the Acoustical Society of Korea
• /
• v.28 no.5
• /
• pp.441-446
• /
• 2009

This paper proposes an automatic indexing algorithm of golf video using audio information. In the proposed algorithm, the input audio stream is demultiplexed into the stream of video and audio. By means of Adaboost-cascade classifier, the continuous audio stream is classified into announcer's speech segment recorded in studio, music segment accompanied with players' names on TV screen, reaction segment of audience according to the play, reporter's speech segment with field background, filed noise segment like wind or waves. And golf swing sound including drive shot, iron shot, and putting shot is detected by the method of impulse onset detection and modulation spectrum verification. The detected swing and applause are used effectively to index action or highlight unit. Compared with video based semantic analysis, main advantage of the proposed system is its small computation requirement so that it facilitates to apply the technology to embedded consumer electronic devices for fast browsing.

• The Journal of the Acoustical Society of Korea
• /
• v.29 no.3
• /
• pp.209-215
• /
• 2010

Recently portable terminal is received attention by wireless networks and mass capacity ROM. In this result, TTS(Text to Speech) system is inserted to portable terminal. Nevertheless high quality synthesis is difficult in portable terminal, users need high quality synthesis. In this paper, we proposed Distributed TTS (DTTS) that was composed of server and terminal. The DTTS on corpus based speech synthesis can be high quality synthesis. Synthesis system in server that generate optimized speech concatenation information after database search and transmit terminal. Synthesis system in terminal make high quality speech synthesis as low computation using transmitted speech concatenation information from server. The proposed method that can be reducing complexity, smaller power consumption and efficient maintenance.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.3
• /
• pp.140-147
• /
• 2024

In the Republic of Korea, people with hearing impairments are the second-largest demographic within the registered disability community, following those with physical disabilities. Despite this demographic significance, research on sign language translation technology is limited due to several reasons including the limited market size and the lack of adequately annotated datasets. Despite the difficulties, a few researchers continue to improve the performacne of sign language translation technologies by employing the recent advance of deep learning, for example, the transformer architecture, as the transformer-based models have demonstrated noteworthy performance in tasks such as action recognition and video classification. This study focuses on enhancing the recognition performance of sign language translation by combining transformers with 3D-CNN. Through experimental evaluations using the PHOENIX-Wether-2014T dataset [1], we show that the proposed model exhibits comparable performance to existing models in terms of Floating Point Operations Per Second (FLOPs).

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.4
• /
• pp.29-33
• /
• 2024

Recent advancements in mobile apps have led to continuously increasing memory demands on smartphone systems. Unlike desktops, which use swap functions to backup the entire memory footprint to storage when memory space is exhausted, smartphones terminate apps and lose significant context. This occurs because large-scale I/O operations to flash memory cause severe delays when swap is enabled on smartphones. This paper discusses how efficient memory management can be performed by using eMRAM, which is faster in write operations than flash memory, as the swap area in mobile systems. Considering the characteristics of backup storage (i.e., flash memory for the file system and eMRAM for the swap areas) as well as the reference characteristics of each page, we demonstrate that the proposed memory allocation and reclamation policies can improve the smartphone's I/O performance by an average of 15%.

• Journal of Advanced Navigation Technology
• /
• v.28 no.4
• /
• pp.428-435
• /
• 2024

In this paper, a human activity regeneration (HAR) system based on multiple input multiple output frequency modulation continuous wave (MIMO FMCW) radar was designed and implemented. Using point cloud data from MIMO radar sensors has advantages in terms of privacy, safety, and accuracy. For the implementation of the HAR system, a customized neural network based on PointPillars and depthwise separate convolutional neural network (DS-CNN) was developed. By processing high-resolution point cloud data through a lightweight network, high accuracy and efficiency were achieved. As a result, the accuracy of 98.27% and the computational complexity of 11.27M multiply-accumulates (Macs) were achieved. In addition, the developed neural network model was implemented on Raspberry-Pi embedded system and it was confirmed that point cloud data can be processed at a speed of up to 8 fps.

• Information Systems Review
• /
• v.26 no.2
• /
• pp.83-93
• /
• 2024

This study proposes a preprocessing methodology based on Dynamic Time Warping (DTW) and k-Nearest Neighbors (kNN) to effectively represent time series data for predicting the completion quality of electroplating baths. The proposed DTW-based kNN preprocessing approach was applied to various regression models and compared. The results demonstrated a performance improvement of up to 43% in maximum RMSE and 24% in MAE compared to traditional decision tree models. Notably, when integrated with neural network-based regression models, the performance improvements were pronounced. The combined structure of the proposed preprocessing method and regression models appears suitable for situations with long time series data and limited data samples, reducing the risk of overfitting and enabling reasonable predictions even with scarce data. However, as the number of data samples increases, the computational load of the DTW and kNN algorithms also increases, indicating a need for future research to improve computational efficiency.

• Journal of Broadcast Engineering
• /
• v.22 no.4
• /
• pp.484-495
• /
• 2017

The development of display devices and the increase of network transmission bandwidth bring demands for over 2K high resolution video such as panorama video, 4K ultra-high definition commercial broadcasting, and ultra-wide viewing video. To compress these image sequences with significant amount of data, High Efficiency Video Coding (HEVC) standard with the highest coding efficiency is a promising solution. HEVC, the latest video coding standard, provides high encoding efficiency using various advanced encoding tools, but it also requires significant amounts of computation complexity compared to previous coding standards. In particular, the complexity of HEVC decoding process is a imposing challenges on real-time playback of ultra-high resolution video. To accelerate the HEVC decoding process for ultra high resolution video, this paper introduces a data-level parallel video decoding method using slice and/or tile supported by HEVC. Moreover, deblocking filter process is further parallelized. The proposed method distributes independent decoding operations of each tile and/or each slice to multiple threads as well as deblocking filter operations. The experimental results show that the proposed method facilitates executions up to 2.0 times faster than the HEVC reference software for 4K videos.

• Journal of Broadcast Engineering
• /
• v.20 no.1
• /
• pp.92-109
• /
• 2015

Broadcast encryption is a cryptographic primitive that allows a sender to securely broadcast a message to a set of receivers. The most influential broadcast encryption system was proposed in 2001 by Naor, Naor, Lotspiech, based on a pseudo-random generator and the Subset Difference (SD) method. In this paper, we suggest a new broadcast encryption system that is based on secret sharing and SD methods. On an efficiency aspect, our system achieves O(r) transmission cost, O($log^2n$) storage cost, and O(1) computational cost for the number n of users and the number r of revoked users. Compared to O(log n) computational cost in the previous SD method, our system has the advantage that it needs only constant-sized computational cost for decryption, regardless of the number n or r. On a security aspect, our system can achieve tighter security reduction than the previous SD method and the gap of security loss is about O(n log n). Moreover, our result shows that it is possible to give the effect of the SD method while using an information-theoretically secure key distribution technique as in the Complete Subtree method.

• Journal of KIISE:Software and Applications
• /
• v.36 no.6
• /
• pp.451-461
• /
• 2009

The Adaptive Random Testing(ART) aims to improve the performance of traditional Random Testing(RT) by reducing the number of test cases to find the failure region which is located in the input domain. Such enhancement can be obtained by efficient selection algorithms of test cases. The ART through Iterative Partitioning(IP-ART) is one of ART techniques and it uses an iterative input domain partitioning method to improve the performance of early-versions of ART which have significant drawbacks in computation time. And the IP-ART with Enlarged Input Domain(EIP-ART), an improved version of IP-ART, is known to make additional performance improvement with scalability by expanding to virtual test space beyond real input domain of IP-ART. The EIP-ART algorithm, however, have the drawback of heavy cost of computation time to generate test cases mainly due to the virtual input domain enlargement. For this reason, two algorithms are proposed in this paper to mitigate the computation overhead of the EIP-ART. In the experiments by simulations, the tiling technique of input domain, one of two proposed algorithms, showed significant improvements in terms of computation time and testing performance.

• Journal of Korea Spatial Information System Society
• /
• v.7 no.1 s.13
• /
• pp.25-37
• /
• 2005

Recently, with the development of wireless communications and mobile computing, interest about mobile computing is rising. Mobile computing can be regarded as an environment where a user carries mobile devices, such as a PDA or a notebook, and shares resources with a server computer via wireless communications. A mobile database refers to a database which is used in these mobile devices. The mobile database can be used in the fields of insurance business, banking business, medical treatment, and so on. Especially, LBS(Location Based Service) which utilizes location information of users becomes an essential field of mobile computing. In order to support LBS in the mobile environment, there must be an Embedded Spatial MMDBMS(Main-Memory Database Management System) that can efficiently manage large spatial data in spatial mobile devices. Therefore, in this paper, we designed and implemented the Embedded Spatial MMDBMS, extended from the HSQLDB which is an existing MMDBMS for PC, to manage spatial data efficiently in spatial mobile devices. The Embedded Spatial MMDBMS adopted the spatial data model proposed by ISO(International Organization for Standardization), provided the arithmetic coding method that is suitable for spatial data, and supported the efficient spatial index which uses the MBR compression and hashing method suitable for spatial mobile devices. In addition, the system offered the spatial data display capability in low-performance processors of spatial mobile devices and supported the data caching and synchronization capability for performance improvement of spatial data import/export between the Embedded Spatial MMDBMS and the GIS server.