• Smart Structures and Systems
• /
• v.24 no.6
• /
• pp.745-757
• /
• 2019

Road surface deteriorations such as potholes have caused motorists heavy monetary damages every year. However, effective road condition monitoring has been a continuing challenge to road owners. Depth cameras have a small field of view and can be easily affected by vehicle bouncing. Traditional image processing methods based on algorithms such as segmentation cannot adapt to varying environmental and camera scenarios. In recent years, novel object detection methods based on deep learning algorithms have produced good results in detecting typical objects, such as faces, vehicles, structures and more, even in scenarios with changing object distances, camera angles, lighting conditions, etc. Therefore, in this study, a Deep Learning Pothole Detector (DLPD) based on the deep region-based convolutional neural network is proposed for autonomous detection of potholes from images. About 900 images with potholes and road surface conditions are collected and divided into training and testing data. Parameters of the network in the DLPD are calibrated based on sensitivity tests. Then, the calibrated DLPD is trained by the training data and applied to the 215 testing images to evaluate its performance. It is demonstrated that potholes can be automatically detected with high average precision over 93%. Potholes can be differentiated from manholes by training and applying a manhole-pothole classifier which is constructed using the convolutional neural network layers in DLPD. Repeated detection of the same potholes can be prevented through feature matching of the newly detected pothole with previously detected potholes within a small region.

• Current Optics and Photonics
• /
• v.3 no.1
• /
• pp.8-15
• /
• 2019

We describe a common optical system that merges a LADAR system, which generates a point cloud, and a more traditional imaging system operating in the LWIR, which generates image data. The optimum diameter of the entrance pupil was determined by analysis of detection ranges of the LADAR sensor, and the result was applied to design a common optical system using LADAR sensors and LWIR sensors; the performance of these sensors was then evaluated. The minimum detectable signal of the $128{\times}128-pixel$ LADAR detector was calculated as 20.5 nW. The detection range of the LADAR optical system was calculated to be 1,000 m, and according to the results, the optimum diameter of the entrance pupil was determined to be 15.7 cm. The modulation transfer function (MTF) in relation to the diffraction limit of the designed common optical system was analyzed and, according to the results, the MTF of the LADAR optical system was 98.8% at the spatial frequency of 5 cycles per millimeter, while that of the LWIR optical system was 92.4% at the spatial frequency of 29 cycles per millimeter. The detection, recognition, and identification distances of the LWIR optical system were determined to be 5.12, 2.82, and 1.96 km, respectively.

• Journal of Korea Multimedia Society
• /
• v.22 no.2
• /
• pp.136-145
• /
• 2019

The crowded environment of a domestic pig farm is highly vulnerable to the spread of infectious diseases such as foot-and-mouth disease, and studies have been conducted to automatically analyze behavior of pigs in a crowded pig farm through a video surveillance system using a camera. Although it is required to correctly separate occluding pigs for tracking each individual pigs, extracting the boundaries of the occluding pigs fast and accurately is a challenging issue due to the complicated occlusion patterns such as X shape and T shape. In this study, we propose a fast and accurate method to separate occluding pigs not only by exploiting the characteristics (i.e., one of the fast deep learning-based object detectors) of You Only Look Once, YOLO, but also by overcoming the limitation (i.e., the bounding box-based object detector) of YOLO with the test-time data augmentation of rotation. Experimental results with two-pigs occlusion patterns show that the proposed method can provide better accuracy and processing speed than one of the state-of-the-art widely used deep learning-based segmentation techniques such as Mask R-CNN (i.e., the performance improvement over Mask R-CNN was about 11 times, in terms of the accuracy/processing speed performance metrics).

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.12
• /
• pp.4776-4794
• /
• 2020

We propose a rapid adaptive learning framework for streaming object detection, called EER-ASSL. The method combines the expected error reduction (EER) dependent rollback learning and the active semi-supervised learning (ASSL) for a rapid adaptive CNN detector. Most CNN object detectors are built on the assumption of static data distribution. However, images are often noisy and biased, and the data distribution is imbalanced in a real world environment. The proposed method consists of collaborative sampling and EER-ASSL. The EER-ASSL utilizes the active learning (AL) and rollback based semi-supervised learning (SSL). The AL allows us to select more informative and representative samples measuring uncertainty and diversity. The SSL divides the selected streaming image samples into the bins and each bin repeatedly transfers the discriminative knowledge of the EER and CNN models to the next bin until convergence and incorporation with the EER rollback learning algorithm is achieved. The EER models provide a rapid short-term myopic adaptation and the CNN models an incremental long-term performance improvement. EER-ASSL can overcome noisy and biased labels in varying data distribution. Extensive experiments shows that EER-ASSL obtained 70.9 mAP compared to state-of-the-art technology such as Faster RCNN, SSD300, and YOLOv2.

• Journal of radiological science and technology
• /
• v.44 no.4
• /
• pp.375-380
• /
• 2021

This study aim was quantitative assessment of MTFs of spectrum of the square wave chart images and Coltman chart images for 0°, 1.7°, 2.2°, 2.9°, 4.1° by using chart method. In general device was AccuRay-650 (DK Medical System, Korea) used, indirect flat panel detector(FPD) Aero (Konica, Japan) used and MATLAB R2019a (MathWorks, USA) used. The result of comparison for each angle of MTF the edge image was highest quantitatively value for MTF finding of showed the best value of 0.1 based on the frequency of 3.5 mm^-1, value of 0.1 based on the square wave was frequency of 3.0 mm^-1 and value of 0.1 based on the Coltman transform was frequency of 2.4 mm^-1. In this study it was significant that the methodology of the international Electro-technical Commission was applied mutandis by using the Fujita method within 2~3°.

• Journal of Korea Society of Industrial Information Systems
• /
• v.27 no.1
• /
• pp.11-18
• /
• 2022

The use of multi-spectral cameras is essential for day and night pedestrian detection. In this paper, a color camera and a thermal imaging infrared camera were used to detect pedestrians near a crosswalk for 24 hours at an intersection with a high risk of traffic accidents. For pedestrian detection, the YOLOv5 object detector was used, and the detection performance was improved by using color images and thermal images at the same time. The proposed system showed a high performance of 0.940 mAP in the day/night multi-spectral (color and thermal image) pedestrian dataset obtained from the actual crosswalk site.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.16 no.5
• /
• pp.1671-1686
• /
• 2022

Face manipulation tools represented by Deepfake have threatened the security of people's biological identity information. Particularly, manipulation tools with deep learning technology have brought great challenges to Deepfake detection. There are many solutions for Deepfake detection based on traditional machine learning and advanced deep learning. However, those solutions of detectors almost have problems of poor performance when evaluated on different quality datasets. In this paper, for the sake of making high-quality Deepfake datasets, we provide a preprocessing method based on the image pixel matrix feature to eliminate similar images and the residual channel attention network (RCAN) to resize the scale of images. Significantly, we also describe a Deepfake detector named Cascaded-Hop which is based on the PixelHop++ system and the successive subspace learning (SSL) model. By feeding the preprocessed datasets, Cascaded-Hop achieves a good classification result on different manipulation types and multiple quality datasets. According to the experiment on FaceForensics++ and Celeb-DF, the AUC (area under curve) results of our proposed methods are comparable to the state-of-the-art models.

• Journal of Astronomy and Space Sciences
• /
• v.40 no.4
• /
• pp.237-245
• /
• 2023

Laser communication has been considered as a novel method for earth observation satellites with generation of high data volume. It offers faster data transmission speeds compared to conventional radio frequency (RF) communication due to the short wavelength and narrow beam divergence. However, laser beams are refracted due to atmospheric turbulence between the ground and the satellite. Refracted laser beams, upon reaching the receiver, result in angle-of-arrival (AoA) fluctuation, inducing image dancing and wavefront distortion. These phenomena hinder signal acquisition and lead to signal loss in the course of laser communication. So, precise alignment between the transmitter and receiver is essential to guarantee effective and reliable laser communication, which is achieved by pointing, acquisition, and tracking (PAT) system. In this study, we simulate the effectiveness of tip/tilt compensation for more efficient laser communication in the satellite-ground downlink. By compensating for low-order terms using tip/tilt mirror, we verify the alleviation of AoA fluctuations under both weak and strong atmospheric turbulence conditions. And the performance of tip/tilt correction is analyzed in terms of the AoA fluctuation and collected power on the detector.

• Progress in Medical Physics
• /
• v.7 no.1
• /
• pp.65-77
• /
• 1996

Using the 2-D and 3-D Hoffman brain phantom, 3-D Jaszczak phantom and Single Photon Emission Computed Tomography, the effects of data acquisition parameter, attenuation, noise, scatter and reconstruction algorithm on image quantitation as well as image quality were studied. For the data acquisition parameters, the images were acquired by changing the increment angle of rotation and the radius. The less increment angle of rotation resulted in superior image quality. Smaller radius from the center of rotation gave better image quality, since the resolution degraded as increasing the distance from detector to object increased. Using the flood data in Jaszczak phantom, the optimal attenuation coefficients were derived as 0.12cm$\^$-1/ for all collimators. Consequently, the all images were corrected for attenuation using the derived attenuation coefficients. It showed concave line profile without attenuation correction and flat line profile with attenuation correction in flood data obtained with jaszczak phantom. And the attenuation correction improved both image qulity and image quantitation. To study the effects of noise, the images were acquired for 1min, 2min, 5min, 10min, and 20min. The 20min image showed much better noise characteristics than 1min image indicating that increasing the counting time reduces the noise characteristics which follow the Poisson distribution. The images were also acquired using dual-energy windows, one for main photopeak and another one for scatter peak. The images were then compared with and without scatter correction. Scatter correction improved image quality so that the cold sphere and bar pattern in Jaszczak phantom were clearly visualized. Scatter correction was also applied to 3-D Hoffman brain phantom and resulted in better image quality. In conclusion, the SPECT images were significantly affected by the factors of data acquisition parameter, attenuation, noise, scatter, and reconstruction algorithm and these factors must be optimized or corrected to obtain the useful SPECT data in clinical applications.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.2 no.1 s.2
• /
• pp.25-40
• /
• 2003

The purpose of this thesis is development of detection algorithm for stop-line detector. Detail detection area is set in basing detection area($1.8{\times}4.0m$) and traffic information(volume, occupancy, nonoccupancy) is collected by passive infrared detector at designing detection area. The basis detection area($1.8{\times}4.0m$) is named existing PIR and detection area applied on development algorithm is named proposal PIR. The proposal PIR is collected data such volume, occupancy, nonoccupancy, speed and lane change, but this thesis is limited to evaluate for volume, occupancy and nonoccupancy The procedure and each step of being developed algorithm is described in the next (1) The detection area of proposal PIR is made up of 2 of $1.8{\times}0.6m$ size(the detection area is named 1 and 3) and 1 of $1.8{\times}1.78m$ size(the detection area is named 2) (2) The image detection area is set on monitor to analyze outdoor photographing data then video frame analysis has been done by analyzer. (3) The occupancy, nonoccupancy and speed data of vehicle have been collected with the detection area 1 and 3 and lane change has been collected with combination of detection area 1, 2 and 3 The MAD and MAPE have been utilized to being compared with volume, occupancy and nonoccupancy for the field application and evaluation of a algorithm As the result, the proposal PIR data have been identified superior to the existing PIR data and the effect has been improved its information(volume, occupancy and nonoccupancy)