• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.11
• /
• pp.319-328
• /
• 2023

Unlike cloud computing, edge computing technology analyzes and judges data close to devices and users, providing advantages such as real-time service, sensitive data protection, and reduced network traffic. EdgeX Foundry, a representative open source of edge computing platforms, is an open source-based edge middleware platform that provides services between various devices and IT systems in the real world. EdgeX Foundry provides a service for handling camera devices, along with a service for handling existing sensed data, which only supports simple streaming and camera device management and does not store or process image data obtained from the device inside EdgeX. This paper presents a technique that can store and process image data inside EdgeX by applying some of the services provided by EdgeX Foundry. Based on the proposed technique, a service pipeline for object detection services used core in the field of autonomous driving was created for experiments and performance evaluation, and then compared and analyzed with existing methods.

• Transactions on Semiconductor Engineering
• /
• v.2 no.1
• /
• pp.17-20
• /
• 2024

This paper presents a novel lightweight object detection model tailored for low-powered edge devices, addressing the limitations of traditional resource-intensive computer vision models. Our proposed detector, inspired by the Single Shot Detector (SSD), employs a compact yet robust network design. Crucially, it integrates an 'enhancer block' that significantly boosts its efficiency in detecting smaller objects. The model comprises two primary components: the Light_Block for efficient feature extraction using Depth-wise and Pointwise Convolution layers, and the Enhancer_Block for enhanced detection of tiny objects. Trained from scratch on the Udacity Annotated Dataset with image dimensions of 300x480, our model eschews the need for pre-trained classification weights. Weighing only 5.5MB with approximately 0.43M parameters, our detector achieved a mean average precision (mAP) of 27.7% and processed at 140 FPS, outperforming conventional models in both precision and efficiency. This research underscores the potential of lightweight designs in advancing object detection for edge devices without compromising accuracy.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.5
• /
• pp.605-611
• /
• 2023

Accidents at construction sites have a very high rate of fatalities due to the nature of being prone to severe injury patients. In order to reduce the mortality rate of severely injury patients, quick response is required, and some systems that detect accidents using AI technology and cameras have been devised to respond quickly to accidents. However, since existing accident detection systems use only a single camera, there are blind spots, Thus, they cannot detect all accidents at a construction site. Therefore, in this paper, we present the system that minimizes the detection blind spot by using multiple cameras. Our implemented system extracts feature points from the images of multiple cameras with the YOLO-pose library, and inputs the extracted feature points to a Long Short Term Memory-based recurrent neural network in order to detect accidents. In our experimental result, we confirme that the proposed system shows high accuracy while minimizing detection blind spots by using multiple cameras.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.12
• /
• pp.3383-3397
• /
• 2023

Scene graphs serve as semantic abstractions of images and play a crucial role in enhancing visual comprehension and reasoning. However, the performance of Scene Graph Generation is often compromised when working with biased data in real-world situations. While many existing systems focus on a single stage of learning for both feature extraction and classification, some employ Class-Balancing strategies, such as Re-weighting, Data Resampling, and Transfer Learning from head to tail. In this paper, we propose a novel approach that decouples the feature extraction and classification phases of the scene graph generation process. For feature extraction, we leverage a transformer-based architecture and design an adaptive calibration function specifically for predicate classification. This function enables us to dynamically adjust the classification scores for each predicate category. Additionally, we introduce a Distribution Alignment technique that effectively balances the class distribution after the feature extraction phase reaches a stable state, thereby facilitating the retraining of the classification head. Importantly, our Distribution Alignment strategy is model-independent and does not require additional supervision, making it applicable to a wide range of SGG models. Using the scene graph diagnostic toolkit on Visual Genome and several popular models, we achieved significant improvements over the previous state-of-the-art methods with our model. Compared to the TDE model, our model improved mR@100 by 70.5% for PredCls, by 84.0% for SGCls, and by 97.6% for SGDet tasks.

• Quality Improvement in Health Care
• /
• v.29 no.2
• /
• pp.2-14
• /
• 2023

Purpose: This study aimed to identify strategies to enhance the competencies of patient safety coordinators in Korea. Methods: Fourteen participants from nine hospitals were interviewed between May and November 2022. Qualitative content analysis was used to analyze the data. Results: As for the strategies to enhance patient safety management competency, 3 themes and 11 sub-themes were derived. The first theme was 'Having individual competence as a patient safety coordinator', and the sub-themes were 'Communication skills with members', 'Flexible thinking from multiple perspectives', and 'Preparing for administrative work competencies that they had not experienced as a nurse.' The second theme was 'Responding strategically to promote improvement activities', and the sub-themes for it were 'Multi-angle approach to the problem', 'A careful approach so as not to be taken as criticism in the field', 'Increasing the possibility of improvement activities through awareness', 'Activating the network between patient safety coordinators', and 'Expanding learning opportunities through patient safety case analysis.' The third theme was 'Obtaining support to facilitate patient safety activities', and the sub-themes for this were 'Improving staff awareness of patient safety', 'Providing a training course for nurse professional of patient safety', and 'Expanding the manpower allocation standard of patient safety coordinators.' Conclusion: This study explored personal competencies such as document writing and computer utilization capabilities, focused on ways to improve the field of patient safety management, and emphasized the need for organizational and political support.

• Korean Journal of Radiology
• /
• v.21 no.7
• /
• pp.891-899
• /
• 2020

Objective: To assess the diagnostic performance of a deep learning-based algorithm for automated detection of acute and chronic rib fractures on whole-body trauma CT. Materials and Methods: We retrospectively identified all whole-body trauma CT scans referred from the emergency department of our hospital from January to December 2018 (n = 511). Scans were categorized as positive (n = 159) or negative (n = 352) for rib fractures according to the clinically approved written CT reports, which served as the index test. The bone kernel series (1.5-mm slice thickness) served as an input for a detection prototype algorithm trained to detect both acute and chronic rib fractures based on a deep convolutional neural network. It had previously been trained on an independent sample from eight other institutions (n = 11455). Results: All CTs except one were successfully processed (510/511). The algorithm achieved a sensitivity of 87.4% and specificity of 91.5% on a per-examination level [per CT scan: rib fracture(s): yes/no]. There were 0.16 false-positives per examination (= 81/510). On a per-finding level, there were 587 true-positive findings (sensitivity: 65.7%) and 307 false-negatives. Furthermore, 97 true rib fractures were detected that were not mentioned in the written CT reports. A major factor associated with correct detection was displacement. Conclusion: We found good performance of a deep learning-based prototype algorithm detecting rib fractures on trauma CT on a per-examination level at a low rate of false-positives per case. A potential area for clinical application is its use as a screening tool to avoid false-negative radiology reports.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.24 no.1
• /
• pp.17-22
• /
• 2024

This paper introduces the implementation of a secondhand clothing trading system equipped with virtual fitting functionality based on deep learning. The proposed system provides users with the ability to visually try on secondhand clothing items online and assess their fit. To achieve this, it utilizes the Convolutional Neural Network (CNN) algorithm to create virtual representations of users considering their body shape and the design of the clothing. This enables buyers to pre-assess the fit of clothing items online before actually wearing them, thereby aiding in their purchase decisions. Additionally, sellers can present accurate clothing sizes and fits through the system, enhancing customer satisfaction. This paper delves into the CNN model's training process, system implementation, user feedback, and validates the effectiveness of the proposed system through experimental results.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.19 no.1
• /
• pp.189-196
• /
• 2024

Frequent false positives alarm from the Intelligent Selective Control System have raised significant concerns. These persistent issues have led to declines in operational efficiency and market credibility among agents. Developing a new model or replacing the existing one to mitigate false positives alarm entails substantial opportunity costs; hence, improving the quality of the training dataset is pragmatic. However, smaller organizations face challenges with inadequate capabilities in dataset collection and refinement. This paper proposes an automatic human pose data collection system centered around a human pose estimation model, utilizing camera-based sensor fusion techniques and edge devices. The system facilitates the direct collection and real-time processing of field data at the network periphery, distributing the computational load that typically centralizes. Additionally, by directly labeling field data, it aids in constructing new training datasets.

• Journal of the Society of Disaster Information
• /
• v.20 no.1
• /
• pp.117-127
• /
• 2024

Purpose: The purpose is to conduct basic research to save energy and improve the safety of manufacturing plant infrastructure by comprehensively monitoring energy management, temperature, humidity, dust and gas, air quality, and machine operation status in small and medium-sized manufacturing plants. Method: To this end, energy-related data and environmental information were collected in real time through digital power meters and IoT sensors, and research was conducted to disseminate and respond to situations for energy saving through monitoring and analysis based on the collected information. Result: We presented an application plan that takes into account energy management, cost reduction, and safety improvement, which are key indicators of ESG management activities. Conclusion: This study utilized various sensor devices and related devices in a smart factory as a practical case study in a company. Based on the information collected through research, a basic system for energy saving and safety improvement was presented.

• Journal of Internet Computing and Services
• /
• v.15 no.3
• /
• pp.45-52
• /
• 2014

The Ubiquitous-City (U-City) is a smart or intelligent city to satisfy human beings' desire to enjoy IT services with any device, anytime, anywhere. It is a future city model based on Internet of everything or things (IoE or IoT). It includes a lot of video cameras which are networked together. The networked video cameras support a lot of U-City services as one of the main input data together with sensors. They generate huge amount of video information, real big data for the U-City all the time. It is usually required that the U-City manipulates the big data in real-time. And it is not easy at all. Also, many times, it is required that the accumulated video data are analyzed to detect an event or find a figure among them. It requires a lot of computational power and usually takes a lot of time. Currently we can find researches which try to reduce the processing time of the big video data. Cloud computing can be a good solution to address this matter. There are many cloud computing methodologies which can be used to address the matter. MapReduce is an interesting and attractive methodology for it. It has many advantages and is getting popularity in many areas. Video cameras evolve day by day so that the resolution improves sharply. It leads to the exponential growth of the produced data by the networked video cameras. We are coping with real big data when we have to deal with video image data which are produced by the good quality video cameras. A video surveillance system was not useful until we find the cloud computing. But it is now being widely spread in U-Cities since we find some useful methodologies. Video data are unstructured data thus it is not easy to find a good research result of analyzing the data with MapReduce. This paper presents an analyzing system for the video surveillance system, which is a cloud-computing based video data management system. It is easy to deploy, flexible and reliable. It consists of the video manager, the video monitors, the storage for the video images, the storage client and streaming IN component. The "video monitor" for the video images consists of "video translater" and "protocol manager". The "storage" contains MapReduce analyzer. All components were designed according to the functional requirement of video surveillance system. The "streaming IN" component receives the video data from the networked video cameras and delivers them to the "storage client". It also manages the bottleneck of the network to smooth the data stream. The "storage client" receives the video data from the "streaming IN" component and stores them to the storage. It also helps other components to access the storage. The "video monitor" component transfers the video data by smoothly streaming and manages the protocol. The "video translator" sub-component enables users to manage the resolution, the codec and the frame rate of the video image. The "protocol" sub-component manages the Real Time Streaming Protocol (RTSP) and Real Time Messaging Protocol (RTMP). We use Hadoop Distributed File System(HDFS) for the storage of cloud computing. Hadoop stores the data in HDFS and provides the platform that can process data with simple MapReduce programming model. We suggest our own methodology to analyze the video images using MapReduce in this paper. That is, the workflow of video analysis is presented and detailed explanation is given in this paper. The performance evaluation was experiment and we found that our proposed system worked well. The performance evaluation results are presented in this paper with analysis. With our cluster system, we used compressed $1920{\times}1080(FHD)$ resolution video data, H.264 codec and HDFS as video storage. We measured the processing time according to the number of frame per mapper. Tracing the optimal splitting size of input data and the processing time according to the number of node, we found the linearity of the system performance.