• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.9
• /
• pp.1330-1339
• /
• 2022

3D object detection generally aims to detect relatively large data such as automobiles, buses, persons, furniture, etc, so it is vulnerable to small object detection. In addition, in an environment with limited resources such as embedded devices, it is difficult to apply the model because of the huge amount of computation. In this paper, the accuracy of small object detection was improved by focusing on local features using only one layer, and the inference speed was improved through the proposed knowledge distillation method from large pre-trained network to small network and adaptive quantization method according to the parameter size. The proposed model was evaluated using SUN RGB-D Val and self-made apple tree data set. Finally, it achieved the accuracy performance of 62.04% at mAP@0.25 and 47.1% at mAP@0.5, and the inference speed was 120.5 scenes per sec, showing a fast real-time processing speed.

• Maxillofacial Plastic and Reconstructive Surgery
• /
• v.32 no.1
• /
• pp.37-42
• /
• 2010

Purpose: The mental foramen (MF) is an important anatomical structure during local anesthesia and surgical procedures in terms of achieving effective mental nerve blocks and avoiding injuries to the neurovascular bundles. Thus, understanding the anatomic features of the mandibular canal and accessory mental foramen in Korean could contribute to the surgical anatomic assessment. This study was to elucidate frequency, position and course of AMF (accessory mental foramen) in Korean using 3D cone beam computed tomography. Materials and Methods: The CBCT (Conbeam computed tomography) DICOM data (Alphard, Asahi, Japan) from 540 patients in korean were analyzed. We investigated images of 3D CBCT using Ondemand (CyberMed, Korea) software program on the incidence and anatomical characteristics of accessory foramen. Results: The accessory mental foramina were found in 17 patients. Accessory mental foramina exist predominantly in the apical area of the second premolar and posteroinferior area of the mental foramen. The accessory branches of the mandibular canal showed common characteristics in the course of gently sloping posterosuperior direction in the buccal surface area. The size of most AMF was obviously smaller than that of MF. Conclusion: We could identify frequency, position and course of AMF (accessory mental foramen) by the anatomical study of the accessory mental foramen using 3D cone beam CT in Korean.

• Journal of the Semiconductor & Display Technology
• /
• v.15 no.2
• /
• pp.61-65
• /
• 2016

Facial feature detection is a fundamental function in the field of computer vision such as security, bio-metrics, 3D modeling, and face recognition. There are many algorithms for the function, active shape model is one of the most popular local texture models. This paper addresses issues related to face detection, and implements an efficient extraction algorithm for extracting the facial feature points to use on iOS platform. In this paper, we extend the original ASM algorithm to improve its performance by four modifications. First, to detect a face and to initialize the shape model, we apply a face detection API provided from iOS CoreImage framework. Second, we construct a weighted local structure model for landmarks to utilize the edge points of the face contour. Third, we build a modified model definition and fitting more landmarks than the classical ASM. And last, we extend and build two-dimensional profile model for detecting faces within input images. The proposed algorithm is evaluated on experimental test set containing over 500 face images, and found to successfully extract facial feature points, clearly outperforming the original ASM.

• Korean Journal of Mineralogy and Petrology
• /
• v.33 no.1
• /
• pp.1-10
• /
• 2020

The atomic structures of silicate liquids at high pressure provide insights into the transport properties including thermal conductivities or elemental partitioning behavior between rocks and magmas in Earth's interior. Whereas the local electronic structure around silicon may vary with the arrangement of the nearby oxygens, the detailed nature of such relationship remains to be established. Here, we explored the atomic origin of the pressure-induced changes in the electronic structure around silicon by calculating the partial electronic density of states and L₃-edge X-ray absorption spectra of SiO₂ polymorphs. The result showed that the Si PDOS at the conduction band varies with the crystal structure and local atomic environments. Particularly, d-orbital showed the distinct features at 108 and 130 eV upon the changes in the coordination number of Si. Calculated Si XAS spectra showed features due to the s,d-orbitals at the conduction band and varied similarly with those observed in s,d-orbitals upon changes in the crystal structures. The calculated Si XAS spectrum for α-quartz was analogous to the experimental Si XRS spectrum for SiO₂ glass, implying the overall similarities in the local atomic environments around the Si. The edge energies at the center of gravity of XAS spectra were closely related to the Si-O distance, thus showing the systematic changes upon densification. Current results suggest that the Si L_2,3-edge XRS, sensitive probe of the Si-O distance, would be useful in unveiling the densification mechanism of silicate glasses and melts at high pressure.

• Journal of Mechanical Science and Technology
• /
• v.19 no.12
• /
• pp.2213-2223
• /
• 2005

In order to reconstruct a full 3D human model in reverse engineering (RE), a 3D scanner needs to be placed arbitrarily around the target model to capture all part of the scanned surface. Then, acquired multiple scans must be registered and merged since each scanned data set taken from different position is just given in its own local co-ordinate system. The goal of the registration is to create a single model by aligning all individual scans. It usually consists of two sub-steps: rough and fine registration. The fine registration process can only be performed after an initial position is approximated through the rough registration. Hence an automated rough registration process is crucial to realize a completely automatic RE system. In this paper an automated rough registration method for aligning multiple scans of complex human face is presented. The proposed method automatically aligns the meshes of different scans with the information of features that are extracted from the estimated principal curvatures of triangular meshes of the human face. Then the roughly aligned scanned data sets are further precisely enhanced with a fine registration step with the recently popular Iterative Closest Point (ICP) algorithm. Some typical examples are presented and discussed to validate the proposed system.

• Tunnel and Underground Space
• /
• v.33 no.6
• /
• pp.508-518
• /
• 2023

This paper presents a new approach for the automatic mapping of discontinuities in a tunnel face based on its 3D digital model reconstructed by LiDAR scan or photogrammetry techniques. The main idea revolves around the identification of discontinuity areas in the 3D digital model of a tunnel face by segmenting its 2D projected images using a deep-learning semantic segmentation model called U-Net. The proposed deep learning model integrates various features including the projected RGB image, depth map image, and local surface properties-based images i.e., normal vector and curvature images to effectively segment areas of discontinuity in the images. Subsequently, the segmentation results are projected back onto the 3D model using depth maps and projection matrices to obtain an accurate representation of the location and extent of discontinuities within the 3D space. The performance of the segmentation model is evaluated by comparing the segmented results with their corresponding ground truths, which demonstrates the high accuracy of segmentation results with the intersection-over-union metric of approximately 0.8. Despite still being limited in training data, this method exhibits promising potential to address the limitations of conventional approaches, which only rely on normal vectors and unsupervised machine learning algorithms for grouping points in the 3D model into distinct sets of discontinuities.

• Journal of the Institute of Convergence Signal Processing
• /
• v.23 no.4
• /
• pp.234-240
• /
• 2022

This paper proposes a customized AI exercise recommendation service for balancing the relative amount of exercise according to the working environment by each occupation. WISDM database is collected by using acceleration and gyro sensors, and is a dataset that classifies physical activities into 18 categories. Our system recommends a adaptive exercise using the analyzed activity type after classifying 18 physical activities into 3 physical activities types such as whole body, upper body and lower body. 1 Dimensional convolutional neural network is used for classifying a physical activity in this paper. Proposed model is composed of a convolution blocks in which 1D convolution layers with a various sized kernel are connected in parallel. Convolution blocks can extract a detailed local features of input pattern effectively that can be extracted from deep neural network models, as applying multi 1D convolution layers to input pattern. To evaluate performance of the proposed neural network model, as a result of comparing the previous recurrent neural network, our method showed a remarkable 98.4% accuracy.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.3
• /
• pp.251-256
• /
• 2019

In this paper we proposed system and algorithm to recognize hand gestures based on the millimeter wave that is in 65GHz bandwidth. The proposed system is composed of millimeter wave radar board, analog to data conversion and data capture board and notebook to perform gesture recognition algorithms. As feature vectors in proposed algorithm. we used global and local zernike moment descriptor which are robust to distort by rotation of scaling of 2D data. As Experimental result, performance of the proposed algorithm is evaluated and compared with those of algorithms using single global or local zernike descriptor as feature vectors. In analysis of confusion matrix of algorithms, the proposed algorithm shows the better performance in comparison of precision, accuracy and sensitivity, subsequently total performance index of our method is 95.6% comparing with another two mehods in 88.4% and 84%.

• Journal of the Korean Magnetics Society
• /
• v.3 no.1
• /
• pp.48-55
• /
• 1993

We report the photoemission (PES) studies for the Co/Pd multilayter. The Co 3d PES spectrum of Co/Pd exhibits two interesting features, one near the Fermi energy, $E_{F}$, and another at ~2.5 eV below $E_{F}$. The Co 3d peak near $E_{F}$ of Co/Pd is much narrower than that of the bulk Co, consistent with the enhanced Co magnetic moment in Co/Pd compared to that in the bulk Co. The Co 3d feature at ~-2.5 eV resembles the Pd valence band structures, which suggests a substantial hybridization between the Co and Pd sublayers. The Co 3d PES spectrum of Co/Pd is compared with the existing band structures, obtained using the local spin density functional calculations. A reasonable agreement is found concerning the bandwidth of the occupied part of the Co 3d band, whereas a narrow Co 3d peak near $E_{F}$ seems not to be described by the band structure calculations.

• The Journal of Engineering Geology
• /
• v.11 no.2
• /
• pp.205-214
• /
• 2001

A simple automated measuring method for planar or linear features on the rock excavation surface is presented. Attitude of the planar and linear feature is calculated from 3D coordinates of points on the structures. Spatial coordinates are calculated from overlapped stereo images. Factors used in the calculation are (1) local coordinates of the left and right images, (2) the focal length of cameras, and (3) the distance between two cameras. A simple image capturing device and an image treatment routine coded by Visual Basic and GIS components are constructed for the remote measurements, The methodology shows less than 1 cm error when a point is measured from 179 cm in distance. The methodology is tested at the excavation site in PaiChai University. Remotely measured result matches well with the manual measurement within the reasonable error range.