• Journal of the Semiconductor & Display Technology
• /
• v.22 no.2
• /
• pp.128-132
• /
• 2023

Recently, with the development of deep learning technology, research on pedestrian attribute recognition technology using deep neural networks has been actively conducted. Existing pedestrian attribute recognition techniques can be obtained in such a way as global-based, regional-area-based, visual attention-based, sequential prediction-based, and newly designed loss function-based, depending on how pedestrian attributes are detected. It is known that the performance of these pedestrian attribute recognition technologies varies greatly depending on the type of backbone network that constitutes the deep neural networks model. Therefore, in this paper, several backbone networks are applied to the baseline pedestrian attribute recognition model and the performance changes of the model are analyzed. In this paper, the analysis is conducted using Resnet34, Resnet50, Resnet101, Swin-tiny, and Swinv2-tiny, which are representative backbone networks used in the fields of image classification, object detection, etc. Furthermore, this paper analyzes the change in time complexity when inferencing each backbone network using a CPU and a GPU.

• Journal of Korean Society for Quality Management
• /
• v.49 no.3
• /
• pp.327-340
• /
• 2021

Purpose: The purpose of this study is to propose a system predicting whether an electricity distribution system is abnormal by analyzing the temperature of the deteriorated system. Traditional electricity distribution system abnormality diagnosis was mainly limited to post-inspection. This research presents a remote monitoring system for detecting thermal images of the deteriorated electricity distribution system efficiently hereby providing safe and efficient abnormal diagnosis to electricians. Methods: In this study, an object detection algorithm (YOLOv5) is performed using 16,866 thermal images of electricity distribution systems provided by KEPCO(Korea Electric Power Corporation). Abnormality/Normality of the extracted system images from the algorithm are classified via the limit temperature. Each classification model, Random Forest, Support Vector Machine, XGBOOST is performed to explore 463,053 temperature datasets. The process capability index is employed to indicate the quality of the electricity distribution system. Results: This research performs case study with transformers representing the electricity distribution systems. The case study shows the following states: accuracy 100%, precision 100%, recall 100%, F1-score 100%. Also the case study shows the process capability index of the transformers with the following states: steady state 99.47%, caution state 0.16%, and risk state 0.37%. Conclusion: The sum of caution and risk state is 0.53%, which is higher than the actual failure rate. Also most transformer abnormalities can be detected through this monitoring system.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2017.06a
• /
• pp.36-39
• /
• 2017

2014 년 Ian Goodfellow 가 발표한 한편의 논문은 머신러닝 분야에 새로운 방향을 제시하였다. Generative Adversarial Networks, 일명 GAN 이라 불리는 이 논문은 이전까지 딥러닝으로 하지못했던 새로운 것을 창조해내는 작업을 하는 첫번째 딥러닝 알고리즘이다. 이전까지는 딥러닝을 통해 영상에서 객체의 종류를 판단하는 Classification 문제나, 영상에서 특정 객체를 검출하여 위치를 찾는 Object detection, 영상 내 특정 객체만 분리해내는 Image segmentation 문제를 해결하고 있었다. GAN 의 등장으로, 다양한 방면에서 GAN 을 적용하여 기존에는 하지 못했던 새로운 분야에 딥러닝을 적용한 사례들이 등장하고 있다. 본 논문에서는 GAN 의 원리 분석과 GAN 을 응용하여 여러 분야에 적용한 사례들을 살펴보고자 한다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2001.10b
• /
• pp.478-480
• /
• 2001

본 논문에서는 실제 도로 영상에서 움직이는 물체를 검출하고 판별하기 위한 새로운 방법을 제안한다. 제안된 방법은 연속된 영상의 차영상에 적응적 임계간을 적용하여 움직임이 있는 후보 영역을 검출한다. 검출된 후보영역에 관심의 대상이 되는 물체의 포함 여부를 판별하기 위해 신경망을 사용한다. 신경망의 입력으로 사용되는 특징 벡터들의 차원을 줄이기 위해, 후보 영역의 스케일 공간 웨이블릿 특징 벡터 (scale-space wavelet feature vector)들을 사용한다. 제안된 방법은 비디오 기반의 응용 프로그램에 유용하게 이용될 수 있으며 특히, 시간에 따라 조명이 변하거나 잡음이 포함된 비디오 영상에 대해 좋은 결과를 얻을 수 있다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.07a
• /
• pp.513-515
• /
• 2015

최근 지능형 차량과 ADAS(Advanced Driver Assistance System) 개발에 있어 차량 영상을 이용한 도로 정보 분석이 중요한 화두로 떠오르고 있다. 다양한 도로 정보 중에서 도로표지판 검출 및 판단은 차량 운행 환경을 파악할 수 있는 중요한 과정이 될 수 있다. 이에 본 논문에서는 차량 영상에서의 색상 정보를 이용하여 표지판의 후보 영역을 추출(Candidate Generation)하고, 후보 영상에 대한 HOG(Histogram of Gradient) 특징 분석을 통해 도로표지판 여부와 그 종류를 판단(Object Classification)하는 알고리즘을 구현하였다. 또한 구현 알고리즘은 실시간 처리가 가능한 속도를 보여주어 지능형 차량 또는 ADAS에서의 실제 적용이 가능하도록 하였다.

• International conference on construction engineering and project management
• /
• 2024.07a
• /
• pp.1308-1308
• /
• 2024

Computer vision techniques have been widely employed in automated construction management to enhance safety and prevent accidents at construction sites. However, previous research in the field of vision-based approaches has often overlooked small-sized construction tools. These tools present unique challenges in data collection due to their diverse shapes and sizes, as well as in improving model performance to accurately detect and classify them. To address these challenges, this study aimed to enhance the performance of vision-based classifiers for small-sized construction tools, including bucket, cord reel, hammer, and tacker, by leveraging synthetic images generated from a 3D virtual environment. Three classifiers were developed using the YOLOv8 algorithm, each differing in the composition of the training dataset: (i) 'Real-4000', trained on 4,000 authentic images collected through web crawling methods (1,000 images per object); (ii) 'Hybrid-4000', consisting of 2,000 authentic images and 2,000 synthetic images; and (iii) 'Hybrid-8000', incorporating 4,000 authentic images and 4,000 synthetic images. To validate the performance of the classifiers, 144 directly-captured images for each object were collected from real construction sites as the test dataset. The mean Average Precision at an IoU threshold of 0.5 (mAP_0.5) for the classifiers was 79.6%, 90.8%, and 94.8%, respectively, with the 'Hybrid-8000' model demonstrating the highest performance. Notably, for objects with significant shape variations, the use of synthetic images led to the enhanced performance of the vision-based classifiers. Moreover, the practical applicability of the proposed classifiers was validated through confidence scores, particularly between the 'Hybrid-4000' and 'Hybrid-8000' models. Statistical analysis using t-tests indicated that the performance of the 'Hybrid-4000' model would either matched or exceeded that of the 'Hybrid-8000'model based on confidence scores. Thus, employing the 'Hybrid-4000' model may be preferable in terms of data collection efficiency and processing time, contributing to enhanced safety and real-time automation and robotics in construction practices.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.5
• /
• pp.361-367
• /
• 2016

The intruder detection system using digital PIR sensor has the problem that it can't recognize human correctly. In this paper, we suggest a new intruder detection system based on analog PIR sensor to get around the drawbacks of the digital PIR sensor. The analog type PIR sensor emits the voltage output at various levels whereas the output of the digitial PIR sensor is binary. The signal captured using analog PIR sensor is sampled, and its frequency feature is extracted using FFT or MFCC. The extracted features are used for the input of neural networks. After neural network is trained using various human and pet's intrusion data, it is used for classifying human and pet in the intrusion situation.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2021.05a
• /
• pp.169-171
• /
• 2021

Image processing through cameras such as self-driving, CCTV, mobile phone security, and parking facilities is being used to solve many real-life problems. Simple classification is solved through image processing, but it is difficult to find images or in-image features of complexly mixed objects. To solve this feature point, we utilize deep learning techniques in classification, detection, and segmentation of image data so that we can think and judge closely. Of course, the results are better than just image processing, but we confirm that the results judged by the method of image segmentation using deep learning have deviations from the real object. In this paper, we study how to perform accuracy improvement through simple image processing just before outputting the output of deep learning image segmentation to increase the precision of image segmentation.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.28 no.3
• /
• pp.10-18
• /
• 2024

Vehicle loading is one of the main causes of bridge deterioration. Although WiM (Weigh in Motion) can be used to measure vehicle loading on a bridge, it has disadvantage of high installation and maintenance cost due to its contactness. In this study, a non-contact method is proposed to estimate the vehicle loading history of bridges using deep learning and CCTV images. The proposed method recognizes the vehicle type using an object detection deep learning model and estimates the vehicle loading based on the load-based vehicle type classification table developed using the weights of empty vehicles of major domestic vehicle models. Faster R-CNN, an object detection deep learning model, was trained using vehicle images classified by the classification table. The performance of the model is verified using images of CCTVs on actual bridges. Finally, the vehicle loading history of an actual bridge was obtained for a specific time by continuously estimating the vehicle loadings on the bridge using the proposed method.

• International conference on construction engineering and project management
• /
• 2020.12a
• /
• pp.313-322
• /
• 2020

The recognition of the risk hazards is a vital step to effectively prevent accidents on a construction site. The advanced development in computer vision systems and the availability of the large visual database related to construction site made it possible to take quick action in the event of human error and disaster situations that may occur during management supervision. Therefore, it is necessary to analyze the risk factors that need to be managed at the construction site and review appropriate and effective technical methods for each risk factor. This research focuses on analyzing Occupational Safety and Health Agency (OSHA) related to risk zone identification rules that can be adopted by the image recognition technology and classify their risk factors depending on the effective technical method. Therefore, this research developed a pattern-oriented classification of OSHA rules that can employ a large scale of safety hazard recognition. This research uses joint reasoning of risk zone Identification and numeric input by utilizing a stereo camera integrated with an image detection algorithm such as (YOLOv3) and Pyramid Stereo Matching Network (PSMNet). The research result identifies risk zones and raises alarm if a target object enters this zone. It also determines numerical information of a target, which recognizes the length, spacing, and angle of the target. Applying image detection joint logic algorithms might leverage the speed and accuracy of hazard detection due to merging more than one factor to prevent accidents in the job site.