• Annual Conference of KIPS
• /
• 2020.05a
• /
• pp.468-471
• /
• 2020

본 논문은 ESRGAN(Enhanced Super Resolution GAN)과 Semantic Soft Segmentation을 이용한 객체 분할의 성능 개선에 관한 것이다. 본 논문의 연구진이 이미 제안한 Mask R-CNN과 Semantic Soft Segmentation을 이용한 객체 분할 방법은 전반적으로 객체 분할 성능이 양호한 반면, 객체의 크기가 상대적으로 작으면 분할 성능이 저조해지는 문제점이 있었다. 본 논문은 이러한 문제점을 해결하기 위한 것으로, Mask R-CNN을 통해 검출된 객체의 크기가 일정 기준치 이하인 경우, ESRGAN을 통해 초해상화를 수행한 후, Semantic Soft Segmentation을 수행함으로써 소형 객체의 분할 성능을 개선함에 그 목적이 있다. 제안된 방법에 따르면, 기존의 방볍에 비해 크기가 작은 객체의 분할 특성을 좀 더 효과적으로 개선할 수 있음을 확인할 수 있었다.

• Journal of Internet Computing and Services
• /
• v.23 no.5
• /
• pp.47-58
• /
• 2022

In recent years, navigation aids, which plays as sea traffic lights, have been digitized, and are developing beyond simple sign purpose to provide various functions such as marine information collection, supervision, control, etc. For example, Busan Port which is located in South Korea is leading the application of the advanced technologies by installing cameras on buoys and recording video images to supervise maritime accidents. However, there are difficulties to perform their major functions since the advanced technologies require long-term battery operation and also management and maintenance of them are hampered by marine characteristics. This study proposes a system that can automatically notify maritime objects passing around buoys by analyzing image information. In the existing sensor-based accident prevention systems, the alarms are generated by a collision detection sensor. The system can identify the cause of the accident whilst even though it is difficult not possible to fundamentally prevent the accidents. Therefore, in order to overcome these limitations, the proposed a maritime object detection system is based on marine characteristics. The experiments demonstrate that the proposed system shows about 5 times faster processing speed than other existing algorithms.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.01a
• /
• pp.99-101
• /
• 2022

이상 행동을 탐지하는 딥러닝 기반 검지 시스템은 동영상 기반 데이터로부터 움직임을 보이는 객체를 추적하고 그 객체의 행동을 분석하여 정상적인 행동 범위를 벗어나는 패턴을 보이는 영역을 이상으로 탐지한다. 특히 생성적 적대 신경망(GAN)과 광학 흐름 추정(Optical flow estimation) 기법을 활용하여 움직임에 대한 특징 정보를 추출하고 이를 학습하여 행동 패턴에 대한 모델링을 수행한다. 모델 학습 및 테스트에 활용되는 데이터셋의 해상도가 낮거나 이상 행동을 표현하는 특징 정보가 부족할 경우 최종 모델 성능에 부정적 영향을 미치게 되며, 특히 광학 흐름이 표현하는 이동량 측면에서 차이가 크게 나지 않는 이상 객체의 경우 탐지가 정확하게 이뤄지지 않는다. 본 연구에서는 동영상 프레임에서 나타나는 객체의 평균 종횡비를 구하고 정상적인 비율을 벗어나는 객체에 대해서 이상 행동을 취하는 샘플로 처리하는 후처리단 모듈을 제안하여 최종적인 모델 성능을 향상시키는 방법을 고안한다.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.9
• /
• pp.21-32
• /
• 2022

Image labeling must be preceded in order to perform object detection, and this task is considered a significant burden in building a deep learning model. Tens of thousands of images need to be trained for building a deep learning model, and human labelers have many limitations in labeling these images manually. In order to overcome these difficulties, this study proposes a method to perform object detection without significant performance degradation, even though labeling some images rather than the entire image. Specifically, in this study, low-resolution oriental painting images are converted into high-quality images using a super-resolution algorithm, and the effect of SSIM and PSNR derived in this process on the mAP of object detection is analyzed. We expect that the results of this study can contribute significantly to constructing deep learning models such as image classification, object detection, and image segmentation that require efficient image labeling.

• The Journal of the Acoustical Society of Korea
• /
• v.37 no.1
• /
• pp.12-20
• /
• 2018

Efforts have been made to reconstruct low-resolution underwater images to high-resolution ones by using the image SR (Super-Resolution) method, all to improve efficiency when acquiring side-scan sonar images. As side-scan sonar images are similar with the optical images with respect to exploiting 2-dimensional signals, conventional image restoration methods for optical images can be considered as a solution. One of the most typical super-resolution methods for optical image is a sparse coding and there are studies for verifying applicability of sparse coding method for underwater images by analyzing sparsity of underwater images. Sparse coding is a method that obtains recovered signal from input signal by linear combination of dictionary and sparse coefficients. However, it requires huge computational load to accurately estimate sparse coefficients. In this study, a sparse coding based underwater image super-resolution method is applied while a selective reconstruction method for object region is suggested to reduce the processing time. For this method, this paper proposes an edge detection and object and non object region classification method for underwater images and combine it with sparse coding based image super-resolution method. Effectiveness of the proposed method is verified by reducing the processing time for image reconstruction over 32 % while preserving same level of PSNR (Peak Signal-to-Noise Ratio) compared with conventional method.

• Journal of KIISE:Software and Applications
• /
• v.29 no.1_2
• /
• pp.114-125
• /
• 2002

In this paper, we are proposing a hierarchical segmentation method that first segments the video data into units of shots by detecting cut and dissolve, and then decides types of camera operations or object movements in each shot. In our previous work[1], each picture group is divided into one of the three detailed categories, Shot(in case of scene change), Move(in case of camera operation or object movement) and Static(in case of almost no change between images), by analysing DC(Direct Current) component of I(Intra) frame. In this process, we have designed two-stage hierarchical neural network with inputs of various multiple features combined. Then, the system detects the accurate shot position, types of camera operations or object movements by searching P(Predicted), B(Bi-directional) frames of the current picture group selectively and hierarchically. Also, the statistical distributions of macro block types in P or B frames are used for the accurate detection of cut position, and another neural network with inputs of macro block types and motion vectors method can reduce the processing time by using only DC coefficients of I frames without decoding and by searching P, B frames selectively and hierarchically. The proposed method classified the picture groups in the accuracy of 93.9-100.0% and the cuts in the accuracy of 96.1-100.0% with three different together is used to detect dissolve, types of camera operations and object movements. The proposed types of video data. Also, it classified the types of camera movements or object movements in the accuracy of 90.13% and 89.28% with two different types of video data.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.8 no.2
• /
• pp.213-218
• /
• 2013

In this paper, a video based ship detection method is proposed to monitor port efficiently. Visual saliency map algorithm and mean shift algorithm is applied to detect moving ships don't include background information which is difficult to track moving ships. It is easy to detect ships at the port using saliency map algorithm, because it is very effective to extract saliency object from background. To remove background information in the saliency region, image segmentation and clustering using mean shift algorithm is used. As results of detecting simulation with images of a camera installed at the harbor, it is shown that the proposed method is effective to detect ships.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.10
• /
• pp.77-83
• /
• 2020

The technologies of 3D spatial information, such as Smart City and Digital Twins, are developing rapidly for managing land and solving urban problems scientifically. In this construction of 3D spatial information, an object using aerial photo images is built as a digital DB. Realistically, the task of extracting a texturing image, which is an actual image of the object wall, and attaching an image to the object wall are important. On the other hand, occluded areas occur in the texturing image. In this study, the ResNet algorithm in deep learning technologies was tested to solve these problems. A dataset was constructed, and the street tree was detected using the ResNet algorithm. The ability of the ResNet algorithm to detect the street tree was dependent on the brightness of the image. The ResNet algorithm can detect the street tree in an image with side and inclination angles.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.40 no.5
• /
• pp.381-391
• /
• 2022

This paper is intended to find one of the prevailing deep learning models that are a type of AI (Artificial Intelligence) that helps rapidly detect damaged buildings where disasters occur. The models selected are SSD-512, RetinaNet, and YOLOv3 which are widely used in object detection in recent years. These models are based on one-stage detector networks that are suitable for rapid object detection. These are often used for object detection due to their advantages in structure and high speed but not for damaged building detection in disaster management. In this study, we first trained each of the algorithms on xBD dataset that provides the post-disaster imagery with damage classification labels. Next, the three models are quantitatively evaluated with the mAP(mean Average Precision) and the FPS (Frames Per Second). The mAP of YOLOv3 is recorded at 34.39%, and the FPS reached 46. The mAP of RetinaNet recorded 36.06%, which is 1.67% higher than YOLOv3, but the FPS is one-third of YOLOv3. SSD-512 received significantly lower values than the results of YOLOv3 on two quantitative indicators. In a disaster situation, a rapid and precise investigation of damaged buildings is essential for effective disaster response. Accordingly, it is expected that the results obtained through this study can be effectively used for the rapid response in disaster management.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.17 no.5
• /
• pp.358-366
• /
• 2024

This paper proposes a method for the real-time camera tracking which detects and employs feature points located on a planar object in 3D space. The proposed approach operates in two stages. First, multiple feature points are detected in the 3D space, and then only those that exist on the planar object are selected. The camera's extrinsic parameters are then estimated using the projective geometry relationship between the feature points of the plane and the camera's image plane. The experiments are conducted in a typical indoor environment with regular lighting, without any special illumination setups. In contrast to conventional approaches, the proposed method can detect new feature points on the planar object in real-time and employ them for the camera tracking. This allows for continuous tracking even when the reference features for the camera pose initialization are not available. The experimental results show an average re-projection error of about 5 to 7 pixels, which is relatively small given the image resolution, and demonstrating that camera tracking is possible even in the absence of reference features within the image.