• Journal of the Korea Society of Computer and Information
• /
• v.25 no.1
• /
• pp.55-61
• /
• 2020

With the development of deep learning technology and advances in computing power, video-based research is now gaining more and more attention. Video data contains a large amount of temporal and spatial information, which is the biggest difference compared with image data. It has a larger amount of data. It has attracted intense attention in computer vision. Among them, motion recognition is one of the research focuses. However, the action recognition of human in the video is extremely complex and challenging subject. Based on many research in human beings, we have found that artificial intelligence-like attention mechanisms are an efficient model for cognition. This efficient model is ideal for processing image information and complex continuous video information. We introduce this attention mechanism into video action recognition, paying attention to human actions in video and effectively improving recognition efficiency. In this paper, we propose a new 3D residual attention network using convolutional neural network based on two attention models to identify human action behavior in the video. An evaluation result of our model showed up to 90.7% accuracy.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.1
• /
• pp.39-47
• /
• 2023

In this paper, we propose a method to measure the similarity between golf swings in videos. As it is known that deep learning-based artificial intelligence technology is effective in the field of computer vision, attempts to utilize artificial intelligence in video-based sports data analysis are increasing. In this study, the joint coordinates of a person in a golf swing video were obtained using a deep learning-based pose estimation model, and based on this, the similarity of each swing segment was measured. For the evaluation of the proposed method, driver swing videos from the GolfDB dataset were used. As a result of measuring swing similarity by pairing swing videos of a total of 36 players, 26 players evaluated that their other swing sequence was the most similar, and the average ranking of similarity was confirmed to be about 5th. This ensured that the similarity could be measured in detail even when the motion was performed similarly.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.3 no.1
• /
• pp.38-46
• /
• 2010

This paper describes a computer vision system based on active IR illumination for real-time gaze discrimination system. Unlike most of the existing gaze discrimination techniques, which often require assuming a static head to work well and require a cumbersome calibration process for each person, our gaze discrimination system can perform robust and accurate gaze estimation without calibration and under rather significant head movement. This is made possible by a new gaze calibration procedure that identifies the mapping from pupil parameters to screen coordinates using generalized regression neural networks (GRNNs). With GRNNs, the mapping does not have to be an analytical function and head movement is explicitly accounted for by the gaze mapping function. Furthermore, the mapping function can generalize to other individuals not used in the training. To further improve the gaze estimation accuracy, we employ a reclassification scheme that deals with the classes that tend to be misclassified. This leads to a 10% improvement in classification error. The angular gaze accuracy is about $5^{\circ}$horizontally and $8^{\circ}$vertically. The effectiveness of our gaze tracker is demonstrated by experiments that involve gaze-contingent interactive graphic display.

• Journal of Korea Multimedia Society
• /
• v.16 no.1
• /
• pp.108-118
• /
• 2013

Recently the 3D modeling techniques were developed rapidly due to rapid development of computer vision, computer graphics with the excellent performance of hardware. With the advent of a variety of 3D contents, 3D modeling technology becomes more in demand and it's quality is increased. 3D face models can be applied widely to such contents with high usability. In this paper, a 3D face modeling is attempted from a given single 2D frontal face image. To achieve the goal, we thereafter the feature points using AAM are extracted from the input frontal face image. With the extracted feature points we deform the 3D general model by 2-pass mesh warping, and also the depth extraction based on intensity values is attempted to. Throughout those processes, a universal 3D face modeling method with less expense and less restrictions to application environment was implemented and it's validity was shown through experiments.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.9
• /
• pp.525-531
• /
• 2019

Efforts to utilize human gestures to effectively implement a more natural and interactive interface between humans and computers have been ongoing in recent years. In this paper, we propose a new algorithm that accepts consecutive three-dimensional (3D) depth images, defines a hand model, and robustly extracts the human hand region based on six palm joints and 15 finger joints. Then, the 3D depth images are adaptively binarized to exclude non-interest areas, such as the background, and accurately extracts only the hand of the person, which is the area of interest. Experimental results show that the presented algorithm detects only the human hand region 2.4% more accurately than the existing method. The hand region extraction algorithm proposed in this paper is expected to be useful in various practical applications related to computer vision and image processing, such as gesture recognition, virtual reality implementation, 3D motion games, and sign recognition.

• Journal of Cadastre & Land InformatiX
• /
• v.52 no.1
• /
• pp.81-93
• /
• 2022

Recently, there have been many studies on object recognition around the vehicle and recognition of traffic signs and traffic lights in autonomous driving. In particular, such the recognition of traffic lights is one of the core technologies in autonomous driving. Therefore, many studies for such the recognition of traffic lights have been performed, the studies based on various deep learning models have increased significantly in recent. In addition, as a high-quality AI training data set for voice, vision, and autonomous driving is released on AIHub, it makes it possible to develop a recognition model for traffic lights suitable for the domestic environment using the data set. In this study, we developed a recognition model for traffic lights that can be used in Korea using the AIHub's training data set. In particular, in order to improve the recognition performance, we used various models of YOLOv4 and YOLOv5, and performed our recognition experiments by defining various classes for the training data. In conclusion, we could see that YOLOv5 shows better performance in the recognition than YOLOv4 and could confirm the reason from the architecture comparison of the two models.

• KIPS Transactions on Software and Data Engineering
• /
• v.2 no.1
• /
• pp.65-70
• /
• 2013

Recently, the field of computer vision has been actively researched through digital image which can be easily generated as the development and expansion of digital camera technology. Especially, research that extracts and utilizes the feature in image has been actively carried out. The image stitching is a method that creates the high resolution image using features extract and match. Image stitching can be widely used in military and medical purposes as well as in variety fields of real life. In this paper, we have proposed efficient image stitching method using fast feature descriptor extraction and matching based on SURF algorithm. It can be accurately, and quickly found matching point by reduction of dimension of feature descriptor. The feature descriptor is generated by classifying of unnecessary minutiae in extracted features. To reduce the computational time and efficient match feature, we have reduced dimension of the descriptor and expanded orientation window. In our results, the processing time of feature matching and image stitching are faster than previous algorithms, and also that method can make natural-looking stitched image.

• Journal of Broadcast Engineering
• /
• v.18 no.1
• /
• pp.77-87
• /
• 2013

In this paper, we present a novel method for local image descriptor called exact order based descriptor (EOD) which is robust to illumination changes and Gaussian noise. Exact orders of image patch is induced by changing discrete intensity value into k-dimensional continuous vector to resolve the ambiguity of ordering for same intensity pixel value. EOD is generated from overall distribution of exact orders in the patch. The proposed local descriptor is compared with several state-of-the-art descriptors over a number of images. Experimental results show that the proposed method outperforms many state-of-the-art descriptors in the presence of illumination changes, blur and viewpoint change. Also, the proposed method can be used for many computer vision applications such as face recognition, texture recognition and image analysis.

• KIPS Transactions on Software and Data Engineering
• /
• v.9 no.6
• /
• pp.177-186
• /
• 2020

Recent advances in deep learning technology have improved image recognition performance in the field of computer vision, and serverless computing is emerging as the next generation cloud computing technology for event-based cloud application development and services. Attempts to use deep learning and serverless computing technology to increase the number of real-world image recognition services are increasing. Therefore, this paper describes how to develop an efficient deep learning based image recognition service system using serverless computing technology. The proposed system suggests a method that can serve large neural network model to users at low cost by using AWS Lambda Server based on serverless computing. We also show that we can effectively build a serverless computing system that uses a large neural network model by addressing the shortcomings of AWS Lambda Server, cold start time and capacity limitation. Through experiments, we confirmed that the proposed system, using AWS Lambda Serverless Computing technology, is efficient for servicing large neural network models by solving processing time and capacity limitations as well as cost reduction.

• Journal of the HCI Society of Korea
• /
• v.10 no.2
• /
• pp.45-55
• /
• 2015

In recent years, keypoint recognition and tracking technologies are considered as crucial task in many practical systems for markerless augmented reality. The keypoint recognition and technologies are widely studied in many research areas, including computer vision, robot navigation, human computer interaction, and etc. Moreover, due to the rapid growth of mobile market related to augmented reality applications, several effective keypoint-based matching and tracking methods have been introduced by considering mobile embedded systems. Therefore, in this paper, we extensively analyze the recent research trends on keypoint-based recognition and tracking with several core components: keypoint detection, description, matching, and tracking. Then, we also present one of our research related to mobile augmented reality, named mobile tour guide system, by real-time recognition and tracking of tour maps on mobile devices.