• Journal of Information Processing Systems
• /
• 제19권4호
• /
• pp.439-449
• /
• 2023

This paper deals with a resource efficient artificial intelligence (AI) service architecture for multi-channel video streams. As an AI service, we consider the object detection model, which is the most representative for video applications. Since most object detection models are basically designed for a single channel video stream, the utilization of the additional resource for multi-channel video stream processing is inevitable. Therefore, we propose a resource efficient AI service framework, which can be associated with various AI service models. Our framework is designed based on the modular architecture, which consists of adaptive frame control (AFC) Manager, multiplexer (MUX), adaptive channel selector (ACS), and YOLO interface units. In order to run only a single YOLO process without regard to the number of channels, we propose a novel approach efficiently dealing with multi-channel input streams. Through the experiment, it is shown that the framework is capable of performing object detection service with minimum resource utilization even in the circumstance of multi-channel streams. In addition, each service can be guaranteed within a deadline.

• Journal of Information Processing Systems
• /
• 제11권1호
• /
• pp.39-56
• /
• 2015

Continuous multi-interval prediction (CMIP) is used to continuously predict the trend of a data stream based on various intervals simultaneously. The continuous integrated hierarchical temporal memory (CIHTM) network performs well in CMIP. However, it is not suitable for CMIP in real-time mode, especially when the number of prediction intervals is increased. In this paper, we propose a real-time integrated hierarchical temporal memory (RIHTM) network by introducing a new type of node, which is called a Zeta1FirstSpecializedQueueNode (ZFSQNode), for the real-time continuous multi-interval prediction (RCMIP) of data streams. The ZFSQNode is constructed by using a specialized circular queue (sQUEUE) together with the modules of original hierarchical temporal memory (HTM) nodes. By using a simple structure and the easy operation characteristics of the sQUEUE, entire prediction operations are integrated in the ZFSQNode. In particular, we employed only one ZFSQNode in each level of the RIHTM network during the prediction stage to generate different intervals of prediction results. The RIHTM network efficiently reduces the response time. Our performance evaluation showed that the RIHTM was satisfied to continuously predict the trend of data streams with multi-intervals in the real-time mode.

• International Journal of Internet, Broadcasting and Communication
• /
• 제7권2호
• /
• pp.168-172
• /
• 2015

In this paper, a balanced bitrate control in transcoding process based on video complexity measure for multi-view system which simultaneously shows multiple channels or video contents in single screen, is proposed. In order to consider the total quality of multiple video streams, the proposed algorithm reduces the complexity of multiple video stream and video quality differences at the same time by controlling bitrates of each stream by weighting when they are stitched for single screen. For the measure of complexity and quality differences between video streams, two different data: histogram of macroblock type and bitrate for each stream are used. The experimental result indicates that proposed algorithm decreases fluctuation of quality difference between videos in the multi-view system.

• 대한기계학회논문집B
• /
• 제31권2호
• /
• pp.195-201
• /
• 2007

A visualization study of flow characteristics in a mixer using multi-nozzle bubbling was performed. The mixer is filed with liquid glycerin (dynamic viscosity = $1000mPa{\cdot}s\;at\;25^{\circ}C$) and convective mixing is induced by air bubbles generated from 9 orifices installed on the bottom of the mixer. To visualize the flow field, PIV (Particle Image Velocimetry) system consisting of 532nm Nd:YAG laser, $2k{\times}2k$ CCD camera and synchronizer is adopted. The bubbles generated with uniform size and frequency form bubble stream and bubble streams rise vertically without interaction between bubble streams. Mixing efficiency is affected by the height of bubbler and the effective height of bubbler is 20nm from the bottom of the mixer.

• 한국환경과학회지
• /
• 제21권9호
• /
• pp.1115-1129
• /
• 2012

The objective of this study was to analyze long-term annual and seasonal trends of water chemistry on landuse patterns and seasonal precipitation using 72 sampling sites within Namhan River watershed during 2001-2010. Water quality, based on multi-parameters of water temperature(WT), dissolved oxygen(DO), biochemical oxygen demand(BOD), chemical oxygen demand(COD), suspended solids(SS), total nitrogen(TN), total phosphorus(TP), and electric conductivity(EC) varied largely depending on monsoon rain and landuse patterns such as forest, cropland, and residence. Concentrations of BOD and COD as an indicator for organic matter pollution, increased during summer monsoon season at the cropland and residential streams. Values of TN and TP were higher in residential streams than in the forest and cropland streams. In the meantime, DO values had weak relations to the landuse patterns of forest and cropland cover. Water quality was worst in cropland and residential streams, and also most degradated in 4th order streams. Overall, our results suggest that efficient water quality management is required in the cropland and residential landuse streams.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2008년도 추계학술발표대회
• /
• pp.413-416
• /
• 2008

In the presence of multiple continuous queries, multi-query optimizing is a new challenge to process multiple stream data in real-time. So, in this paper, we proposed an approach to optimize multi-query of sliding window on network traffic data streams and do some comparisons to traditional queries without optimizing. We also detail some method of scheduling on different data streams, while different scheduling made different results. We test the results on variety of multi-query processing schedule, and proofed the proposed method is effectively optimized the data stream similar multi-queries.

• 인터넷정보학회논문지
• /
• 제13권3호
• /
• pp.93-105
• /
• 2012

데이터 스트림이란 빠르게 연속적으로 무한히 발생하는 데이터 집합을 의미한다. 최근 다양한 산업의 발달로 인해 이러한 스트림 데이터의 효율적인 처리를 위한 요구 사항들이 늘어나고 있다. 특히 많은 연산 비용을 요구하는 조인 연산의 효율적인 처리는 데이터 스트림 관리 시스템의 성능 향상에 많은 영향을 미친다. 본 논문에서는 다중 조인 연속질의의 효율적인 처리를 위하여 최종 질의 결과에 포함되지 않는 불필요한 중간 조인 결과들을 사전에 제거함으로써 조인 연산의 비용을 감소시키는 방법을 제안한다. 이를 위해 스트림 데이터의 모니터링을 위한 매트릭스 기반의 구조체를 제안하고, 제안된 구조체를 이용한 매트릭스 연산을 통하여 최종 조인 결과의 투플 수를 예측함과 동시에 불필요한 중간 결과들을 만들어내는 투플들을 찾아낸다. 이를 통해 해당 투플을 이용한 조인 연산의 수행 여부를 결정하여 최종 조인 결과를 만들지 않는 투플을 조인 연산에서 배제함으로써 효율적으로 다중 조인 연속 질의를 처리한다.

• 대한임베디드공학회논문지
• /
• 제1권2호
• /
• pp.90-103
• /
• 2006

There are many file systems in various operating systems. Those are usually designed for server environments, where the common cases are usually 'multiple active users', 'great many small files' And they assume a big main memory to be used as buffer cache. So the existing file systems are not suitable for resource hungry embedded systems that process multimedia data streams. In this study, we designed and implemented a new file system which efficiently stores and retrieves multimedia data steams. The proposed file system has a very simple disk layout, which guarantees a quick disk initialization and file system recovery. And we introduced a new indexing-scheme, called the time-based indexing scheme, with the file system. With the indexing scheme, the file system maintains the relation between time and the location for all the multimedia streams. The scheme is useful in searching and playing the compressed multimedia streams by locating exact frame position with given time, resulting in reduction of CPU processing and power consumption. The proposed file system and its APIs utilizing the time-based indexing schemes were implemented firstly on a Linux environment, though it is operating system independent. In the performance evaluation on a real DVR system, which measured the execution time of multi-threaded reading and writing, we found the proposed file system is maximum 38.7% faster than EXT2 file system.

• 한국산업정보학회논문지
• /
• 제11권3호
• /
• pp.69-81
• /
• 2006

In this paper, an adaptive algorithm is proposed in streaming MPEG-4 contents with fluctuating resource amount such as throughput of network conditions. In the area of adaptive streaming issue, a lot of researches have been made on how to represent encoded media(such as video) bitstream in scalable way. By contrast, MPEG-4 supports object-based multimedia content which is composed of various types of media streams such as audio, video, image and other graphical elements. Thus, it can be more effective to provide individual media streams in scalable way for streaming object-based content to heterogeneous environment. The proposed method provides the multiple media streams corresponding to an object with different qualities and bit rate in order to support object based scalability to the MPEG-4 content. In addition, an optimal selection of the multiple streams for each object to meet a given constraint is proposed. The selection process is adopted a multiple choice knapsack problem with multi-step selection for the MPEG-4 objects with different scalability levels. The proposed algorithm enforces the optimal selection process to maintain the perceptual qualities of more important objects at the best effort. The experimental results show that the set of selected media stream for presenting objects meets a current transmission condition with more high perceptual quality.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2009년도 춘계학술발표대회
• /
• pp.71-74
• /
• 2009

The data-flow paradigm has been employed in various application areas. It is particularly useful where large data-streams must be processed, for example in video and audio processing, or for scientific visualization. A video-based 3D reconstruction system should process multiple synchronized video streams. The system exhibits many properties that can be targeted using a data-flow approach that is naturally divided into a sequence of processing tasks. In this paper we introduce our concept to apply the data-flow approach to a multi-video 3D reconstruction system.