• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.390-392
• /
• 2003

Video mosaicing techniques have been widely used in virtual reality environments. Especially in GIS field, video mosaics are becoming more and more common in representing urban environments. Such applications mainly use spherical or panoramic mosaics that are based on images taken from a rotating camera around its nodal point. The viewpoint, however, is limited to location within a small area. On the other hand, 2D-mosaics, which are based on images taken from a translating camera, can acquire data in wide area. The 2D-mosaics still have some problems : it can‘t be applied to images taken from a rotational camera in large angle. To compensate those problems , we proposed a novel method for creating video mosaics in 3D space. The proposed algorithm consists of 4 steps: feature -based optical flow detection, camera orientation, 2D-image projection, and image registration in 3D space. All of the processes are fully automatic and successfully implemented and tested with real images.

• Journal of IKEEE
• /
• v.12 no.1
• /
• pp.51-58
• /
• 2008

This paper presents a new method for building detection and reconstruction from aerial images. In our approach, we extract the useful building location information from the generated disparity map to obtain the segmentation of interested objects and thus reduce significantly unnecessary line segment extracted in low level feature extraction step. Hypothesis selection is carried out by using undirected graph in which close cycles represent complete rooftops hypotheses, and hypothesis are finally tested to contruct building model. We test the proposed method with synthetic images generated from Avenches dataset of Ascona aerial images. The experiment result shows that the extracted 3D line segments of the buildings can be efficiently used for the task of building detection and reconstruction from aerial images.

• Journal of Digital Convergence
• /
• v.11 no.7
• /
• pp.209-214
• /
• 2013

Improve the recognition performance of speech recognition systems as a method for recognizing human listening skills were incorporated into the system. In noisy environments by separating the speech signal and noise, select the desired speech signal. but In terms of practical performance of speech recognition systems are factors. According to recognized environmental changes due to noise speech detection is not accurate and learning model does not match. In this paper, to improve the speech recognition feature extraction using gamma tone and learning model using acoustic model was proposed. The proposed method the feature extraction using auditory scene analysis for human auditory perception was reflected In the process of learning models for recognition. For performance evaluation in noisy environments, -10dB, -5dB noise in the signal was performed to remove 3.12dB, 2.04dB SNR improvement in performance was confirmed.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 1998.06b
• /
• pp.291-300
• /
• 1998

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.2
• /
• pp.729-748
• /
• 2021

Vocal detection is one of the fundamental steps in musical information retrieval. Typically, the detection process consists of feature extraction and classification steps. Recently, neural networks are shown to outperform traditional classifiers. In this paper, we report our study on how to improve detection accuracy further by carefully choosing the parameters of the deep network model. Through experiments, we conclude that a feature-classifier model is still better than an end-to-end model. The recommended model uses a spectrogram as the input plane and the classifier is an 18-layer convolutional neural network (CNN). With this arrangement, when compared with existing literature, the proposed model improves the accuracy from 91.8% to 94.1% in Jamendo dataset. As the dataset has an accuracy of more than 90%, the improvement of 2.3% is difficult and valuable. If even higher accuracy is required, the ensemble learning may be used. The recommend setting is a majority vote with seven proposed models. Doing so, the accuracy increases by about 1.1% in Jamendo dataset.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.562-565
• /
• 2009

This paper describes an automatic 3-dimensional (3D) segmentation method for 3D CT (Computed Tomography) images using region growing (RG) and edge detection techniques. Specifically, an augmented RG method in which the contours of regions are extracted by a 3D digital edge detection filter is presented. The feature of this method is the capability of preventing the leakage of regions which is a defect of conventional RG method. Experimental results applied to the extraction of teeth from 3D CT data of jaw bones show that teeth are correctly extracted by the proposed method.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.55 no.4
• /
• pp.173-180
• /
• 2006

A key component of autonomous navigation of intelligent home robot is localization and map building with recognized features from the environment. To validate this, accurate measurement of relative location between robot and features is essential. In this paper, we proposed relative localization algorithm based on 3D reconstruction of scale invariant features of two images which are captured from two parallel cameras. We captured two images from parallel cameras which are attached in front of robot and detect scale invariant features in each image using SIFT(scale invariant feature transform). Then, we performed matching for the two image's feature points and got the relative location using 3D reconstruction for the matched points. Stereo camera needs high precision of two camera's extrinsic and matching pixels in two camera image. Because we used two cameras which are different from stereo camera and scale invariant feature point and it's easy to setup the extrinsic parameter. Furthermore, 3D reconstruction does not need any other sensor. And the results can be simultaneously used by obstacle avoidance, map building and localization. We set 20cm the distance between two camera and capture the 3frames per second. The experimental results show :t6cm maximum error in the range of less than 2m and ${\pm}15cm$ maximum error in the range of between 2m and 4m.

• Journal of the Society of Naval Architects of Korea
• /
• v.59 no.4
• /
• pp.235-242
• /
• 2022

As ship condition inspection technology has been developed, research on collecting, analyzing, and diagnosing condition information has become active. In ships, related research has been conducted, such as analyzing, detecting, and classifying major hull failures such as cracks and corrosion using 2D and 3D data information. However, for geometric deformation such as indents and bulges, 2D data has limitations in detection, so 3D data is needed to utilize spatial feature information. In this study, we aim to detect hull structural deformation positions. It builds a specimen based on actual hull structure deformation and acquires a point cloud from a model scanned with a 3D scanner. In the obtained point cloud, deformation(outliers) is found with a combination of RANSAC algorithms that find the best matching model in the Octree data structure and dataset.

• Journal of Computational Design and Engineering
• /
• v.2 no.4
• /
• pp.253-260
• /
• 2015

Feature-based NC machining, which requires high quality of 3D CAD model, is widely used in machining aircraft structural part. However, there has been little research on how to automatically detect the CAD model errors. As a result, the user has to manually check the errors with great effort before NC programming. This paper proposes an automatic CAD model errors detection approach for aircraft structural part. First, the base faces are identified based on the reference directions corresponding to machining coordinate systems. Then, the CAD models are partitioned into multiple local regions based on the base faces. Finally, the CAD model error types are evaluated based on the heuristic rules. A prototype system based on CATIA has been developed to verify the effectiveness of the proposed approach.

• KSCE Journal of Civil and Environmental Engineering Research
• /
• v.40 no.4
• /
• pp.437-444
• /
• 2020

A lunar rover's optical camera is used to provide navigation and terrain information in an exploration zone. However, due to the scant presence of atmosphere, the Moon has homogeneous terrain with dark soil. Also, in extreme environments, the rover has limited data storage with low computation capability. Thus, for successful exploration, it is required to examine feature detection and matching methods which are robust to lunar terrain and environmental characteristics. In this research, SIFT, SURF, BRISK, ORB, and AKAZE are comparatively analyzed with lunar terrain images from a lunar rover. Experimental results show that SIFT and AKAZE are most robust for lunar terrain characteristics. AKAZE detects less quantity of feature points than SIFT, but feature points are detected and matched with high precision and the least computational cost. AKAZE is adequate for fast and accurate navigation information. Although SIFT has the highest computational cost, the largest quantity of feature points are stably detected and matched. The rover periodically sends terrain images to Earth. Thus, SIFT is suitable for global 3D terrain map construction in that a large amount of terrain images can be processed on Earth. Study results are expected to provide a guideline to utilize feature detection and matching methods for future lunar exploration rovers.