Recently, deep learning-based image recognition systems have been adopted to various surveillance environments, but most of them are still picture-type object recognition methods, which are insufficient for the long term temporal analysis and high-dimensional situation management. Therefore, we propose a method recognizing the specific dangerous situation generated by human in real-time, and utilizing deep learning-based object analysis techniques. The proposed method uses deep learning-based object detection and tracking algorithms in order to recognize the situations such as 'trespassing', 'loitering', and so on. In addition, human's joint pose data are extracted and analyzed for the emergent awareness function such as 'falling down' to notify not only in the security but also in the emergency environmental utilizations.

Recently, high-performance HMDs (Head-Mounted Display) are becoming wireless due to the growth of virtual reality technology. Accordingly, environmental constraints on the hardware usage are reduced, enabling multiple users to experience virtual reality within a single space simultaneously. Existing multi-user virtual reality platforms use the user's location tracking and motion sensing technology based on vision sensors and active markers. However, there is a decrease in immersion due to the problem of overlapping markers or frequent matching errors due to the reflected light. Goal of this study is to develop a multi-user virtual reality moving platform in a single space that can resolve sensing errors and user immersion decrease. In order to achieve this goal hybrid sensing technology was developed, which is the convergence of vision sensor technology for position tracking, IMU (Inertial Measurement Unit) sensor motion capture technology and gesture recognition technology based on smart gloves. In addition, integrated safety operation system was developed which does not decrease the immersion but ensures the safety of the users and supports multimodal feedback. A 6 m×6 m×2.4 m test bed was configured to verify the effectiveness of the multi-user virtual reality moving platform for four users.

In this paper, we propose a classification model based on convolutional neural network(CNN) for predicting 2-year recurrence in non-small cell lung cancer(NSCLC) patients using preoperative chest CT images. Based on the region of interest(ROI) defined as the tumor internal and external area, the input images consist of an intratumoral patch, a peritumoral patch and a peritumoral texture patch focusing on the texture information of the peritumoral patch. Each patch is trained through AlexNet pretrained on ImageNet to explore the usefulness and performance of various patches. Additionally, ensemble learning of network trained with each patch analyzes the performance of different patch combination. Compared with all results, the ensemble model with intratumoral and peritumoral patches achieved the best performance (ACC=98.28%, Sensitivity=100%, NPV=100%).

In this study, character recognition using deep learning is performed among the various defects in the PCB, the purpose of which is to check whether the printed characters are printed correctly on top of components, or the incorrect parts are attached. Generally, character recognition may be perceived as not a difficult problem when considering MNIST, but the printed letters on the PCB component data are difficult to collect, and have very high redundancy. So if a deep learning model is trained with original data without any preprocessing, it can lead to over fitting problems. Therefore, this study aims to reduce the redundancy to the smallest dataset that can represent large amounts of data collected in limited production sites, and to create datasets through data enhancement to train a flexible deep learning model can be used in various production sites. Moreover, ResNet model verifies to determine which combination of datasets is the most effective. This study discusses how to reduce and augment data that is constantly occurring in real PCB production lines, and discusses how to select coresets to learn and apply deep learning models in real sites.

In surgery to remove pancreatic cancer, it is important to figure out the shape of a patient's pancreas. However, previous studies have a limit to detect a pancreas automatically in abdominal CT images, because the pancreas varies in shape, size and location by patient. Therefore, in this paper, we propose a method of learning various shapes of pancreas according to the patients and adjacent slices using Faster R-CNN based on Inception V2, and automatically detecting the pancreas from abdominal CT images. Model training and testing were performed using the NIH Pancreas-CT Dataset, and intensity normalization was applied to all data to improve pancreatic detection accuracy. Additionally, according to the shape of the pancreas, the test dataset was classified into top, middle, and bottom slices to evaluate the model's performance on each data. The results show that the top data's mAP@.50IoU achieved 91.7% and the bottom data's mAP@.50IoU achieved 95.4%, and the highest performance was the middle data's mAP@.50IoU, 98.5%. Thus, we have confirmed that the model can accurately detect the pancreas in CT images.

The worldwide spread of the Internet and the digital information revolution have resulted in a rapid increase in the use and transmission of multimedia information due to the rapid development of communication technologies. It is important to protect images in order to prevent problems such as piracy and illegal distribution. To solve this problem, I propose a new digital color image encryption algorithm in this paper. I design a new pseudo-random number generator based on 1D five-neighborhood maximum length cellular automata (FN-MLCA) to change the pixel values of the plain image into unpredictable values. And then I use a 3D chaotic cat map to effectively shuffle the positions of the image pixel. In this paper, I propose a method to construct a new MLCA by modeling 1D FN-MLCA. This result is an extension of 1D 3-neighborhood CA and shows that more 1D MLCAs can be synthesized. The safety of the proposed algorithm is verified through various statistical analyses.

Due to the large differences in human activity within classes, the large similarity between classes, and the problems of visual angle and occlusion, it is difficult to extract features manually, and the detection rate of human behavior is low. In order to better solve these problems, an improved Faster R-CNN-based detection algorithm is proposed in this paper. It achieves multi-object recognition and localization through a second-order detection network, and replaces the original feature extraction module with Dense-Net, which can fuse multi-level feature information, increase network depth and avoid disappearance of network gradients. Meanwhile, the proposal merging strategy is improved with Soft-NMS, where an attenuation function is designed to replace the conventional NMS algorithm, thereby avoiding missed detection of adjacent or overlapping objects, and enhancing the network detection accuracy under multiple objects. During the experiment, the improved Faster R-CNN method in this article has 84.7% target detection result, which is improved compared to other methods, which proves that the target recognition method has significant advantages and potential.

Aiming Graph convolutional networks (GCNs) have achieved outstanding performances on skeleton-based action recognition. However, several problems remain in existing GCN-based methods, and the problem of low recognition rate caused by single input data information has not been effectively solved. In this article, we propose a Dual-stream fusion method that combines video data and skeleton data. The two networks respectively identify skeleton data and video data and fuse the probabilities of the two outputs to achieve the effect of information fusion. Experiments on two large dataset, Kinetics and NTU-RGBC+D Human Action Dataset, illustrate that our proposed method achieves state-of-the-art. Compared with the traditional method, the recognition accuracy is improved better.

Since the sensor nodes are randomly arranged in the region of interest, it may happen that the sensor network area is separated or there is no sensor node in some area. In addition, after the sensor nodes are deployed in the sensor network, a coverage hole may occur due to the exhaustion of energy or physical destruction of the sensor nodes. The coverage hole can greatly affect the overall performance of the sensor network, such as reducing the data reliability of the sensor network, changing the network topology, disconnecting the data link, and worsening the transmission load. Therefore, sensor network coverage hole recovery has been studied. Existing coverage hole recovery studies present very complex geometric methods and procedures in the two-step process of finding a coverage hole and recovering a coverage hole. This study proposes a method for discovering and recovering a coverage hole in a sensor network, discovering that the sensor node is a boundary node by itself, and determining the location of a mobile node to be added. The proposed method is expected to have better efficiency in terms of complexity and message transmission compared to previous methods.

Smartphones are currently being produced with similar functions, shapes, and software. The foldable smartphone is a product that dramatically changed the shape of the existing smartphone. Therefore, it affects the functions and software. In this paper, we analyze the potential security vulnerability of current mobile authentication methods by dividing them into two parts, security vulnerabilities of non-foldable smartphones, and security vulnerability that appears with the changed smartphone structure. According to the analysis result, the classic and current mobile user authentication methods appears to be easily affected by the smartphone display structure. Finally, we propose an appropriate authentication method as well as the concept of security measures for smartphones with foldable screen. Our method shows that it is more secure than the conventional authentication methods in foldable display smartphone.

In this paper, we propose a method of predicting the next-day stock price fluctuations of 10 KOSDAQ-listed companies in 5G, autonomous driving, and electricity sectors by training SVM, XGBoost, and LightGBM models from macroeconomic·financial market indicators, technical indicators, social interest indicators, and daily positive indices extracted from online news. In the three experiments to find out the usefulness of social interest indicators and daily positive indices, the average accuracy improved when each indicator and index was added to the models. In addition, when feature selection was performed to analyze the superiority of the extracted features, the average importance ranking of the social interest indicator and daily positive index was 5.45 and 1.08, respectively, it showed higher importance than the macroeconomic financial market indicators and technical indicators. With the results of these experiments, we confirmed the effectiveness of the social interest indicators as alternative data and the daily positive index for predicting stock price fluctuation.

Among the formative elements, color and form are the most critical elements that can capture a person's perspective. Companies are concentrating on developing visual perception-oriented products by accommodating the diverse needs of consumers, where the emotional factor plays as an important design consideration. Accordingly, the visual perception reaction of consumers was analyzed through a questionnaire focusing on color and form. The result finds that, when Ernst Gombrich (1909-2001) perceives objects in questionnaires related to personal preferences for color and form, humans do not rely solely on the eyes, but describes with the knowledge diagram through that knowledge. In doing so, the knowledge that we already know plays a part to describe the object. In addition, colors and forms are recognized by combining the social and cultural information experiences of the perceiver with the learned knowledge. In the future studies that define the interrelationships of the formative elements should be continued through the analysis of more complex and clear visual perception characteristics of the formative elements.

Realistic real-time rain streaks rendering has been treated as a very difficult problem because of various natural phenomena. Also, in creating and managing a large number of particles, a large amount of computer resources had to be used. Therefore, in this paper, we propose a more efficient real-time rain streaks rendering algorithm by generating view-dependent rain particles and expressing a large amount of rain even with a small number. By creating a 'rain space' dependent on the field of view of the camera moving in real time, particles are rendered only in that space. Accordingly, even if a small number of particles are rendered, since the rendering is performed in a limited space, an effect of rendering a very large amount of particles can be obtained. This enables very efficient real-time rendering of rain streaks.

Regional brands are recognized for their importance as one of the local public relations policies, but only the methods promoted by local government officials and outside experts are revealing limitations in self-reliance and sustainability. In order for a regional brand to succeed, it is necessary to develop and manage a brand that fits the characteristics of the region together with various stakeholders including residents. In this study, it was confirmed that regional brand evaluation indicators may vary depending on the stage of co-creation in the design process for regional brand development, and the type of value co-creation was determined through a classification method using the results of the evaluation index. According to the results of the study, the type of value co-creation that fits the local situation and characteristics can be derived by adjusting the stage of co-creation execution in the design process.

As service design becomes more common, various user research methods and tools such as cultural probes and contextual inquiry are being used to grasp the user's experience and needs. These methodologies have the advantage of being able to derive various insights through direct observation and investigation, but have disadvantages in that problems such as time and space constraints, cost occurrence, and delay in the service development process may occur. In addition, it is difficult to conduct actual observations, face-to-face investigations, and interviews due to COVID-19, which has recently spread rapidly around the world. Accordingly, prior to the digitization of cultural probes, I would like to suggest the direction of digital tools. First, after grasping the concept of service design and user research through literature research, the necessity of user research in service design was confirmed. After that, the necessity of digitalizing user research was confirmed by extracting functional elements that could be understood and digitalized about the cultural probes methodology, and analyzing previous studies using existing mobile-based survey techniques. Finally, we confirmed the direction of digital tools through user interviews through cultural probes.

In this study, the possibility of contents was explored by extracting materials that can be used as regional-specific contents. To this end, the possibility of local content specialization was examined for Eomuk, the representative food of Busan. First, it looked at the content process of Busan's regional-specific material and developed it as a character. Next, it looked at the story development and the process of incorporating it into animation. In the future, based on this study, we would like to present guidelines for the spread of Korean-style regional-specific content.