• Journal of KIISE:Software and Applications
• /
• v.33 no.10
• /
• pp.851-861
• /
• 2006

The paper proposes a novel unsupervised video object segmentation algorithm for image sequences with low depth-of-field (DOF), which is a popular photographic technique enabling to represent the intention of photographer by giving a clear focus only on an object-of-interest (OOI). The proposed algorithm largely consists of two modules. The first module automatically extracts OOIs from the first frame by separating sharply focused OOIs from other out-of-focused foreground or background objects. The second module tracks OOIs for the rest of the video sequence, aimed at running the system in real-time, or at least, semi-real-time. The experimental results indicate that the proposed algorithm provides an effective tool, which can be a basis of applications, such as video analysis for virtual reality, immersive video system, photo-realistic video scene generation and video indexing systems.

• Journal of IKEEE
• /
• v.24 no.4
• /
• pp.1141-1147
• /
• 2020

The hornet species are so similar in shape that they are difficult for non-experts to classify, and because the size of the objects is small and move fast, it is more difficult to detect and classify the species in real time. In this paper, we developed a system that classifies hornets species in real time based on a deep learning algorithm using a boundary box. In order to minimize the background area included in the bounding box when labeling the training image, we propose a method of selecting only the head and body of the hornet. It also experimentally compares existing boundary box-based object recognition algorithms to find the best algorithms that can detect wasps in real time and classify their species. As a result of the experiment, when the mish function was applied as the activation function of the convolution layer and the hornet images were tested using the YOLOv4 model with the Spatial Attention Module (SAM) applied before the object detection block, the average precision was 97.89% and the average recall was 98.69%.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.43 no.1
• /
• pp.99-107
• /
• 2023

Recently, the rate of death and safety accidents at construction sites is the highest among all kinds of industries. In order to apply artificial intelligence technology to construction sites, it is essential to secure a dataset which can be used as a basic training data. In this paper, a number of image data were collected through actual construction site, for which major construction equipment objects mainly operated in civil engineering sites were defined. The optimal training dataset construction was completed by annotation process of about 90,000 image dataset. Reliability of the dataset was verified with the mAP of over 90 % in use of YOLO, a representative model in the field of object detection. The construction equipment training dataset built in this study has been released which is currently available on the public data portal of the Ministry of Public Administration and Security. This dataset is expected to be freely used for any application of object detection technology on construction sites especially in the field of construction safety in the future.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.08a
• /
• pp.401-405
• /
• 1998

SVAPI 1.0 환경에서의 어구 종속 화자 확인 시스템에 대해 기술한다. 구현된 시스템은 궁극적으로 공중 전화망 응용이 가능한 실용 시스템을 목표로 개발되었으며 이를 위해 SVAPI 위원회에 의해 제안된 SVAPI 1.0을 개발 환경으로 사용하였다. SVAPI는 객체 지향 구조, 클라이언트-서버 및 telephony 환경의 지원등이 특징이며 어플리케이션과 엔진을 독립적으로 개발할 수 있는 이점을 제공한다. 구현된 데모 시스템은 펜티엄 프로세서와 Windows95/NT 4.0 운영체제 그리고 Win16/Win32 API를 통해 제어 가능하며 음성 입력이 가능한 디바이스를 장착한 IBM 호환 PC이다. 화자의 성문 등록은 화자가 동일한 어구를 3회 발성하여 이뤄지며 등록과 확인의 응답속도는 모두 1초 이내이다. 소프트웨어의 구성은 크게 어플리케이션과 어구 종속 화자 확인 엔진으로 구분할 수 있으며 엔진은 끝점 검출 알고리즘, 음성 특징 추출 알고리즘 그리고 연속 HMM 기반의 화자 성문 모델 등록 및 유사도 계산 등을 포함한 확인 알고리즘으로 구성되어 있다. 화자의 성문은이름과 같은 약 3음절 이상의 단어로 등록되고 테스트되었다. 엔진의 객관적인 평가를 위해 전화선을 통해 남자 6명, 여자 3명의 화자로부터 자신의 이름을 각각 40회 발성하여 구축된 음성 데이터 베이스를 사용하였으며 실험 결과 남자는 2.85%, 여자는 2.44%의 EER을 각각 얻었다.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.55 no.2
• /
• pp.52-61
• /
• 2006

For the detection of moving objects, background subtraction methods are widely used. An adaptive Gaussian mixture model combined with probabilistic learning is one of the most popular methods for the real-time update of the complex and dynamic background. However, probabilistic learning approach does not work well in high traffic regions. In this paper, we Propose a reliable learning method of complex and dynamic backgrounds in high traffic regions.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.504-506
• /
• 2022

최근 인공지능 기술의 발달에 따라 여러 분야에 인공지능 기술이 활발히 응용되고 있다. 그중 안전 관리 분야에서 사람 인식을 통한 안전 관리 시스템의 지속적인 개발이 요구되고 있다. 그러나 실내 한정된 공간에서 사람들의 밀집도가 높은 경우 오브젝트의 중복도가 높아져 인식 성능이 낮아질 수 있다. 이를 해결하기 위해 본 논문은 사람의 밀집도가 높은 실내 환경에서 기존 객체 인식 기법의 성능을 분석하였다. 그리고 이러한 제한적인 환경에서 최적의 좋은 성능을 보일 수 있는 SSDLite와 MobileNetV2 모델을 기반으로 soft-NMS 기법을 적용하여 성능을 분석하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.583-586
• /
• 2023

대량 문자서비스를 통한 스팸 문자가 계속 증가하면서 이로 인해 도박, 불법대출 등의 광고성 스팸 문자에 의한 피해가 지속되고 있다. 이러한 문제점을 해결하기 위해 다양한 방법들이 연구되어 왔지만 기존의 방법들은 주로 사전 정의된 키워드나 자주 나오는 단어의 출현 빈도수를 기반으로 스팸 문자를 검출한다. 이는 광고성 문자들이 시스템에서 자동으로 필터링 되는 것을 회피하기 위해 비정상 문자를 조합하여 스팸 문자의 주요 키워드를 의도적으로 변형해 표현하는 경우에는 탐지가 어렵다는 한계가 있다. 따라서, 본 논문에서는 이러한 문제점을 해결하기 위해 딥러닝 기반 객체 탐지 및 OCR 기술을 활용하여 스팸 문자에 사용된 변형된 문자열을 정상 문자열로 복원하고, 변환된 정상 문자열을 문장 수준 이해를 기반으로 하는 자연어 처리 모델을 이용해 스팸 문자 콘텐츠를 분류하는 방법을 제안한다. 그리고 기존 스팸 필터링 시스템에 가장 많이 사용되는 키워드 기반 필터링, 나이브 베이즈를 적용한 방식과의 비교를 통해 성능 향상이 이루어짐을 확인하였다.

• Journal of the Korea institute for structural maintenance and inspection
• /
• v.23 no.7
• /
• pp.80-86
• /
• 2019

As computer performance improves, research using deep learning are being actively carried out in various fields. Recently, deep learning technology has been applying to the safety evaluation for structures. In particular, the internal blades of a turbine structure requires experienced experts and considerable time to detect surface damages because of the difficulty of separation of the blades from the structure and the dark environmental condition. This study proposes a Faster R-CNN deep learning model that can detect surface damages on the internal blades, which is one of the primary elements of the turbine structure. The deep learning model was trained using image data with dent and punch damages. The image data was also expanded using image filtering and image data generator techniques. As a result, the deep learning model showed 96.1% accuracy, 95.3% recall, and 96% precision. The value of the recall means that the proposed deep learning model could not detect the blade damages for 4.7%. The performance of the proposed damage detection system can be further improved by collecting and extending damage images in various environments, and finally it can be applicable for turbine engine maintenance.

• Journal of the Korean Society of Marine Environment & Safety
• /
• v.27 no.1
• /
• pp.22-28
• /
• 2021

This is a basic study on the development of deep learning-based algorithms to detect smoke before the smoke detector operates in the event of a ship fire, analyze and utilize the detected data, and support fire suppression and evacuation activities by predicting the spread of smoke before it spreads to remote areas. Proposed algorithms were reviewed in accordance with the following procedures. As a first step, smoke images obtained through fire simulation were applied to the YOLO (You Only Look Once) model, which is a deep learning-based object detection algorithm. The mean average precision (mAP) of the trained YOLO model was measured to be 98.71%, and smoke was detected at a processing speed of 9 frames per second (FPS). The second step was to estimate the spread of smoke using the coordinates of the boundary box, from which was utilized to extract the smoke geometry from YOLO. This smoke geometry was then applied to the time series prediction algorithm, long short-term memory (LSTM). As a result, smoke spread data obtained from the coordinates of the boundary box between the estimated fire occurrence and 30 s were entered into the LSTM learning model to predict smoke spread data from 31 s to 90 s in the smoke image of a fast fire obtained from fire simulation. The average square root error between the estimated spread of smoke and its predicted value was 2.74.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.33 no.6
• /
• pp.1099-1110
• /
• 2023

The recognition system of autonomous driving and robot navigation performs vision work such as object recognition, tracking, and lane detection after multi-sensor fusion to improve performance. Currently, research on a deep learning model based on the fusion of a camera and a lidar sensor is being actively conducted. However, deep learning models are vulnerable to adversarial attacks through modulation of input data. Attacks on the existing multi-sensor-based autonomous driving recognition system are focused on inducing obstacle detection by lowering the confidence score of the object recognition model.However, there is a limitation that an attack is possible only in the target model. In the case of attacks on the sensor fusion stage, errors in vision work after fusion can be cascaded, and this risk needs to be considered. In addition, an attack on LIDAR's point cloud data, which is difficult to judge visually, makes it difficult to determine whether it is an attack. In this study, image scaling-based camera-lidar We propose an attack method that reduces the accuracy of LCCNet, a fusion model (camera-LiDAR calibration model). The proposed method is to perform a scaling attack on the point of the input lidar. As a result of conducting an attack performance experiment by size with a scaling algorithm, an average of more than 77% of fusion errors were caused.