• 한국방송∙미디어공학회:학술대회논문집
• /
• 한국방송공학회 2009년도 IWAIT
• /
• pp.541-546
• /
• 2009

There are many videos about sports. There is a large need for content based video retrievals. In sports videos, the motions and camera works have much information about shots and plays. This paper proposes the baseball game process understanding using the similar motion retrieval on videos. We can retrieve the similar motion parts based on motions shown in videos using the space-time images describing the motions. Using a finite state model of plays, we can decide the precise point of pitches from the pattern of estimated typical motions. From only the motions, we can decide the precise point of pitches. This paper describes the method and the experimental results.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제9권10호
• /
• pp.4160-4176
• /
• 2015

Data compression like image and video compression has come a long way since the introduction of Compressive Sensing (CS) which compresses sparse signals such as images, videos etc. to very few samples i.e. M < N measurements. At the receiver end, a robust and efficient recovery algorithm estimates the original image or video. Many prominent algorithms solve least squares problem (LSP) iteratively in order to reconstruct the signal hence consuming more processing time. In this paper non-iterative threshold based recovery algorithm (NITRA) is proposed for the recovery of images and videos without solving LSP, claiming reduced complexity and better reconstruction quality. The elapsed time for images and videos using NITRA is in ㎲ range which is 100 times less than other existing algorithms. The peak signal to noise ratio (PSNR) is above 30 dB, structural similarity (SSIM) and structural content (SC) are of 99%.

• 전자통신동향분석
• /
• 제36권2호
• /
• pp.75-82
• /
• 2021

Online action detection (OAD) in a streaming video is an attractive research area that has aroused interest lately. Although most studies for action understanding have considered action recognition in well-trimmed videos and offline temporal action detection in untrimmed videos, online action detection methods are required to monitor action occurrences in streaming videos. OAD predicts action probabilities for a current frame or frame sequence using a fixed-sized video segment, including past and current frames. In this article, we discuss deep learning-based OAD models. In addition, we investigated OAD evaluation methodologies, including benchmark datasets and performance measures, and compared the performances of the presented OAD models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권2호
• /
• pp.713-728
• /
• 2021

This paper proposes an augmented reality technique to generate acrobatic scenes from hitting motion videos. After a user shoots a motion that mimics hitting an object with hands or feet, their pose is analyzed using motion tracking with deep learning to track hand or foot movement while hitting the object. Hitting position and time are then extracted to generate the object's moving trajectory using physics optimization and synchronized with the video. The proposed method can create videos for hitting objects with feet, e.g. soccer ball lifting; fists, e.g. tap ball, etc. and is suitable for augmented reality applications to include virtual objects.

• Journal of Information Processing Systems
• /
• 제15권2호
• /
• pp.344-358
• /
• 2019

Surveillance cameras have installed in many places because security and safety is becoming important in modern society. Through surveillance cameras installed, we can deal with troubles and prevent accidents. However, watching surveillance videos and judging the accidental situations is very labor-intensive. So now, the need for research to analyze surveillance videos is growing. This study proposes an algorithm to track multiple persons using SURF and background subtraction. While the SURF algorithm, as a person-tracking algorithm, is robust to scaling, rotating and different viewpoints, SURF makes tracking errors with sudden changes in videos. To resolve such tracking errors, we combined SURF with a background subtraction algorithm and showed that the proposed approach increased the tracking accuracy. In addition, the background subtraction algorithm can detect persons in videos, and SURF can initialize tracking targets with these detected persons, and thus the proposed algorithm can automatically detect the enter/exit of persons.

• ETRI Journal
• /
• 제44권2호
• /
• pp.327-338
• /
• 2022

Classification models for human action recognition require robust features and large training sets for good generalization. However, data augmentation methods are employed for imbalanced training sets to achieve higher accuracy. These samples generated using data augmentation only reflect existing samples within the training set, their feature representations are less diverse and hence, contribute to less precise classification. This paper presents new data augmentation and action representation approaches to grow training sets. The proposed approach is based on two fundamental concepts: virtual video generation for augmentation and representation of the action videos through robust features. Virtual videos are generated from the motion history templates of action videos, which are convolved using a convolutional neural network, to generate deep features. Furthermore, by observing an objective function of the genetic algorithm, the spatiotemporal features of different samples are combined, to generate the representations of the virtual videos and then classified through an extreme learning machine classifier on MuHAVi-Uncut, iXMAS, and IAVID-1 datasets.

• 한국정보기술학회 영문논문지
• /
• 제10권2호
• /
• pp.125-138
• /
• 2020

Due to the development of camera technology, the cost of producing time-lapse video has been reduced, and time-lapse videos are being applied in many fields. Time-lapse video is created using images obtained by shooting for a long time at long intervals. In this paper, we propose a method to improve the quality of time-lapse videos monitoring the changes in plants. Considering the characteristics of time-lapse video, we propose a method of separating the desired and unnecessary objects and removing unnecessary elements. The characteristic of time-lapse videos that we have noticed is that unnecessary elements appear intermittently in the captured images. In the proposed method, noises are removed by applying a codebook background modeling algorithm to use this characteristic. Experimental results show that the proposed method is simple and accurate to find and remove unnecessary elements in time-lapse videos.

• 한국복식학회:학술대회논문집
• /
• 한국복식학회 2002년도 International Costume Conference
• /
• pp.81-81
• /
• 2002

• 방송공학회논문지
• /
• 제9권2호
• /
• pp.164-173
• /
• 2004

본 논문에서는 등장인물 검출 및 인식과 함께 등장인물의 출연 구간 분석이 가능한 시스템을 제안한다. 드라마, 스포츠와 같은 방송 비디오는 그 특성상 인물이 중심이 되며 각 시점에 등장하는 주요 인물은 방송용 비디오의 중요한 특징이 된다. 본 논문에서는 비디오에서 등장하는 주요 인물을 자동으로 분석하는 ACAV(Automatic Cast-list Analysis in Videos) 시스템을 제안한다. ACAV 시스템은 등장인물을 자동 검출하여 인물 DB에 등록하는 FAGIS(FAce reGIStration)와 생성된 인물 DB을 이용하여 등장인물을 분석하는 FACOG(FAce reCOGnition)로 구성된다. 기존의 상용화된 등장인물 분석 시스템인 FaceIt과의 성능 비교를 통해 ACAV의 성능을 검증하였다. 얼굴 검출 실험에서 ACAV의 얼굴 검출률은 84.3%로 FaceIt 보다 약 30% 높았고, 얼굴 인식 실험에서도 ACAV의 얼굴 인식률은 75.7%로 FaceIt 보다 27.5% 높은 성능을 보였다. ACAV 시스템은 방송 멀티미디어 공급자를 위한 대용량 비디오 관리 시스템으로 이용될 수 있으며 일반 사용자를 대상으로 한 PVR(Personal Video Recorder), 모바일 폰 등의 비디오 관리 시스템으로도 이용될 수 있다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권5호
• /
• pp.2101-2121
• /
• 2020

The rapid growth of instructional videos repositories and their widespread use as a tool to support education have raised the need of studies to assess the quality of those educational resources and their impact on the quality of learning process that depends on them. Khan Academy (KA) repository is one of the prominent educational videos' repositories. It is famous and widely used by different types of learners, students and teachers. To better understand its characteristics and the impact of such repositories on education, we gathered a huge amount of KA data using its API and different web scraping techniques, then we analyzed them. This paper reports the first quantitative and descriptive analysis of Khan Academy repository (KA repository) of open video lessons. First, we described the structure of repository. Then, we demonstrated some analyses highlighting content-based growth and evolution. Those descriptive analyses spotted the main important findings in KA repository. Finally, we focused on users' interactions with video lessons. Those interactions consisted of questions and answers posted on videos. We developed interaction profiles for those videos based on the number of users' interactions. We conducted regression analysis and statistical tests to mine the relation between those profiles and some quality related proposed metrics. The results of analysis showed that all interaction profiles are highly affected by video length and reuse rate in different subjects. We believe that our study demonstrated in this paper provides valuable information in understanding the logic and the learning mechanism inside learning repositories, which can have major impacts on the education field in general, and particularly on the informal learning process and the instructional design process. This study can be considered as one of the first quantitative studies to shed the light on Khan Academy as an open educational resources (OER) repository. The results presented in this paper are crucial in understanding KA videos repository, its characteristics and its impact on education.