• Journal of Information Processing Systems
• /
• 제14권1호
• /
• pp.150-161
• /
• 2018

We propose a novel deep neural network model for detecting human activities in untrimmed videos. The process of human activity detection in a video involves two steps: a step to extract features that are effective in recognizing human activities in a long untrimmed video, followed by a step to detect human activities from those extracted features. To extract the rich features from video segments that could express unique patterns for each activity, we employ two different convolutional neural network models, C3D and I-ResNet. For detecting human activities from the sequence of extracted feature vectors, we use BLSTM, a bi-directional recurrent neural network model. By conducting experiments with ActivityNet 200, a large-scale benchmark dataset, we show the high performance of the proposed DeepAct model.

• 한국멀티미디어학회논문지
• /
• 제19권8호
• /
• pp.1329-1336
• /
• 2016

This paper develops a 360 degree video system using multiple off-the-shelf webcams and a set of embedded boards. Existing 360 degree cameras have shortcomings that they do not support real-time video generation since recorded videos should be copied to computers or smartphones which then provide stitching. Another shortcoming is that wide FoV(Field of View) cameras are not able to provide sufficiently high resolution. Moreover, resulting images are visually distorted bending straight lines. By employing an array of 65 degree FoV webcams, we were able to generate videos on the spot and achieve over 6K resolution with much less distortion. We describe the configuration and algorithms of the proposed system. The performance evaluation results of our early stage prototype system are presented.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2019년도 춘계학술발표대회
• /
• pp.455-457
• /
• 2019

In this paper we study the effect of decomposed spatiotemporal convolutions for action recognition in videos. Our motivation emerges from the empirical observation that spatial convolution applied on solo frames of the video provide good performance in action recognition. In this research we empirically show the accuracy of factorized convolution on individual frames of video for action classification. We take 3D ResNet-18 as base line model for our experiment, factorize its 3D convolution to 2D (Spatial) and 1D (Temporal) convolution. We train the model from scratch using Kinetics video dataset. We then fine-tune the model on UCF-101 dataset and evaluate the performance. Our results show good accuracy similar to that of the state of the art algorithms on Kinetics and UCF-101 datasets.

• 한국컴퓨터정보학회논문지
• /
• 제24권3호
• /
• pp.87-93
• /
• 2019

In this paper, we propose a real-time drone-based violent protest detection system. Our proposed system uses drones to detect scenes of violent protest in real-time. The important problem is that the victims and violent actions have to be manually searched in videos when the evidence has been collected. Firstly, we focused to solve the limitations of existing collecting evidence devices by using drone to collect evidence live and upload in AWS(Amazon Web Service)[1]. Secondly, we built a Deep Learning based violence detection model from the videos using Yolov3 Feature Pyramid Network for human activity recognition, in order to detect three types of violent action. The built model classifies people with possession of gun, swinging pipe, and violent activity with the accuracy of 92, 91 and 80.5% respectively. This system is expected to significantly save time and human resource of the existing collecting evidence.

• International journal of advanced smart convergence
• /
• 제8권2호
• /
• pp.88-93
• /
• 2019

This study explores one of charity drive contents on YouTube channel. Due to the advance of science and technology, ordinary people come to make their own video content online, usually via YouTube. YouTube becomes number one online video storage/streaming platform, and many people upload their own video and they get attention and fame. This study analyzes various aspects of Shoot for Love, soccer-based charity drive videos shown on YouTube channel created in South Korea. Unlike popular videos in YouTube, Shoot for Love centers on charity by casting popular soccer players and celebrities. Especially, this study researches 1) Components 2) Traits of Components 3) Contents of Components in Shoot for Love. Throughout this, it not only analyzes unique aspects of Shoot for Love that show how and why YouTube content matters, but also suggest plausible methods to drive charity and institution are suggested that appeal to the public.

• 한국컴퓨터정보학회논문지
• /
• 제24권6호
• /
• pp.67-72
• /
• 2019

This paper proposes a computer vision-based banknote recognition system intended to assist the blind. This system is robust and fast in recognizing banknotes on videos recorded with a smartphone on real-life scenarios. To reduce the computation time and enable a robust recognition in cluttered environments, this study segments the banknote candidate area from the background utilizing a technique called Pixel-Based Adaptive Segmenter (PBAS). The Speeded-Up Robust Features (SURF) interest point detector is used, and SURF feature vectors are computed only when sufficient interest points are found. The proposed algorithm achieves a recognition accuracy of 98%, a 100% true recognition rate and a 0% false recognition rate. Although Korean banknotes are used as a working example, the proposed system can be applied to recognize other countries' banknotes.

• ETRI Journal
• /
• 제43권3호
• /
• pp.538-548
• /
• 2021

This paper proposes an approach to improve the performance of no-reference video quality assessment for sports videos with dynamic motion scenes using an efficient spatiotemporal model. In the proposed method, we divide the video sequences into video blocks and apply a 3D shearlet transform that can efficiently extract primary spatiotemporal features to capture dynamic natural motion scene statistics from the incoming video blocks. The concatenation of a deep residual bidirectional gated recurrent neural network and logistic regression is used to learn the spatiotemporal correlation more robustly and predict the perceptual quality score. In addition, conditional video block-wise constraints are incorporated into the objective function to improve quality estimation performance for the entire video. The experimental results show that the proposed method extracts spatiotemporal motion information more effectively and predicts the video quality with higher accuracy than the conventional no-reference video quality assessment methods.

• 반도체디스플레이기술학회지
• /
• 제19권4호
• /
• pp.34-38
• /
• 2020

In this paper, we implement a Dashcam system capable of shooting 360 degrees using a Raspberry Pi, shock sensors, distance sensors, and rotating camera with a servo motor. If there is an object approaching the vehicle by the distance sensor, the camera rotates to take a video. In the event of an external shock, videos and images are stored in the server to analyze the cause of the vehicle's accident and prevent the user from forging or tampering with videos or images. We also implement functions that transmit the message with the location and the intensity of the impact when the accident occurs and send the vehicle information to an insurance authority with by linking the system with a smart device. It is advantage that the authority analyzes the transmitted message and provides the accident handling information giving the user's safety and convenience.

• International journal of advanced smart convergence
• /
• 제10권2호
• /
• pp.10-14
• /
• 2021

Today people adopt various contents from their mobile devices which lead to numerous platforms. As technology of 5G, IOT, and smart phone develops, the number of people who create, edit, collect, and share their own videos, photos, and articles continues to increase. As more contents are shared online, the numbers of data being stolen continue to increase too. To prevent these cases, an authentication method is needed to encrypt the content and prove it as its own content. In the report, we propose a few methods to secure various misused content with secondary security. A unique private key is designed when people create new contents through sending photos or videos to platforms. The primary security is to encrypt the "Private Key" with a public key algorithm, making its data-specific "Timeset" that doesn't allow third-party users to enter. For the secondary security, we propose to use Message Authentication Codes(MACs) to certify that we have produced the content.

• 한국멀티미디어학회논문지
• /
• 제24권2호
• /
• pp.186-198
• /
• 2021

Security control using surveillance cameras is established when people observe all surveillance videos directly. However, this task is labor-intensive and it is difficult to detect all abnormal situations. In this paper, we propose a deep neural network model, called AT-Net, that automatically detects abnormal situations in the surveillance video, and introduces an automatic video surveillance system developed based on this network model. In particular, AT-Net alleviates the ambiguity of existing abnormal situation detection methods by mapping features representing relationships between people and objects in surveillance video to the new tensor structure based on sparse coding. Through experiments on actual surveillance videos, AT-Net achieved an F1-score of about 89%, and improved abnormal situation detection performance by more than 25% compared to existing methods.