• IEMEK Journal of Embedded Systems and Applications
• /
• v.16 no.5
• /
• pp.179-186
• /
• 2021

For stereo matching based on deep learning, the design of network structure is crucial to the calculation of matching cost, and the time-consuming problem of convolutional neural network in image processing also needs to be solved urgently. In this paper, a method of stereo matching using sparse loss volume in parallax dimension is proposed. A sparse 3D loss volume is constructed by using a wide step length translation of the right view feature map, which reduces the video memory and computing resources required by the 3D convolution module by several times. In order to improve the accuracy of the algorithm, the nonlinear up-sampling of the matching loss in the parallax dimension is carried out by using the method of multi-category output, and the training model is combined with two kinds of loss functions. Compared with the benchmark algorithm, the proposed algorithm not only improves the accuracy but also shortens the running time by about 30%.

• Spatial Information Research
• /
• v.20 no.4
• /
• pp.29-36
• /
• 2012

The low-carbon freight transportation system was introduced due to increase traffic congestion cost and carbon-dioxide for global climate change according to expanding city logistics demands. It is necessary to create 3D-based road network map for representing realistic road geometry with consideration of fuel consumption and carbon emissions. This study propose that 3D road network model expressed to realistic topography and road structure within trunk road for intercity freight through overlaying 2D-based transport-related thematic map and 1m-resolution DEM. The 3D-based road network map for the experimental road sections(Pyeongtaek harbor-Uiwang IC) was verified by GPS/INS survey and fuel consumption simulation. The results corresponded to effectively reflect realistic road geometry (RMSE=0.87m) except some complex structure such as overpass, and also actual fuel consumption. We expect that Green-based freight route planning and navigation system reflected on 3D geometry of complex road structure will be developed for effectively resolving energy and environmental problems.

• Structural Engineering and Mechanics
• /
• v.49 no.6
• /
• pp.687-703
• /
• 2014

This paper develops a new nonlinear model for active control of three-dimensional (3D) irregular building structures. Both geometrical and material nonlinearities with a neuro-controller training algorithm are applied to a multi-degree-of-freedom 3D system. Two dynamic assembling motions are considered simultaneously in the control model such as coupling between torsional and lateral responses of the structure and interaction between the structural system and the actuators. The proposed control system and training algorithm of the structural system are evaluated by simulating the responses of the structure under the El-Centro 1940 earthquake excitation. In the numerical example, the 3D three-story structure with linear and nonlinear stiffness is controlled by a trained neural network. The actuator dynamics, control time delay and incident angle of earthquake are also considered in the simulation. Results show that the proposed control algorithm for 3D buildings is effective in structural control.

• Journal of IKEEE
• /
• v.16 no.2
• /
• pp.102-108
• /
• 2012

3-dimensional integrated circuits (3D-ICs) have some problems for power delivery network design due to larger supply currents and larger power delivery paths compared to 2D-IC. The power delivery network consists of power bumps & through-silicon-vias (TSVs), and IR-drop at each node varies with the number and location of power bumps & TSVs. It is important to optimize the power bumps & TSVs while IR-drop constraint is satisfied in order to operate chip ordinarily. In this paper, the power bumps & TSVs optimization with optimized power mesh structure for power delivery network in 3D-ICs is proposed.

• Journal of Information Processing Systems
• /
• v.17 no.4
• /
• pp.787-800
• /
• 2021

Three-dimensional (3D) object classification tasks using point clouds are widely used in 3D modeling, face recognition, and robotic missions. However, processing raw point clouds directly is problematic for a traditional convolutional network due to the irregular data format of point clouds. This paper proposes a pointwise convolution neural network (CNN) structure that can process point cloud data directly without preprocessing. First, a 2D convolutional layer is introduced to percept coordinate information of each point. Then, multiple 2D convolutional layers and a global max pooling layer are applied to extract global features. Finally, based on the extracted features, fully connected layers predict the class labels of objects. We evaluated the proposed pointwise CNN structure on the ModelNet10 dataset. The proposed structure obtained higher accuracy compared to the existing methods. Experiments using the ModelNet10 dataset also prove that the difference in the point number of point clouds does not significantly influence on the proposed pointwise CNN structure.

• Proceedings of the KIEE Conference
• /
• 2001.10a
• /
• pp.125-127
• /
• 2001

Switched reluctance motor (SRM) has some advantages such as low cost, high torque density etc. However SRM has inevitably high torque ripple due to the double salient structure. To apply SRM to industrial field, we have to minimize torque ripple, which is the weak-Point of SRM. This paper presents optimal design process of SRM using numerical method such as 2D finite element method (FEM) and 3D equivalent magnetic circuit network method (EMCNM). The electrical and geometrical design parameters have been adopted as 2D design variables. The overhang structure of rotor has been also adopted as 3D design variable. From this work, we can obtain the optimal design, which minimize the torque ripple and maximize energy conversion loop.

• The Journal of Korea Robotics Society
• /
• v.13 no.1
• /
• pp.8-15
• /
• 2018

This paper proposes a convolutional neural network model for distinguishing areas occupied by obstacles from a LiDAR image converted from a 3D point cloud. The channels of a LiDAR image used as input consist of the distances to 3D points, the reflectivities of 3D points, and the heights of 3D points from the ground. The proposed model uses a LiDAR image as an input and outputs a result of a segmented LiDAR image. The proposed model adopts refinement modules with skip connections to segment a LiDAR image. The refinement modules with skip connections in the proposed model make it possible to construct a complex structure with a small number of parameters than a convolutional neural network model with a linear structure. Using the proposed model, it is possible to distinguish areas in a LiDAR image occupied by obstacles such as vehicles, pedestrians, and bicyclists. The proposed model can be applied to recognize surrounding obstacles and to search for safe paths.

• Bulletin of the Korean Chemical Society
• /
• v.35 no.9
• /
• pp.2847-2850
• /
• 2014

• Korean Journal of Materials Research
• /
• v.18 no.3
• /
• pp.163-168
• /
• 2008

The morphology of three-dimensional (3D) cross-linked electrodeposits of copper and tin was investigated as a function of the content of metal sulfate and acetic acid in a deposition bath. The composition of copper sulfate had little effect on the overall copper network structure, whereas that of tin sulfate produced significant differences in the tin network structure. The effect of the metal sulfate content on the copper and tin network is discussed in terms of whether or not hydrogen evolution occurs on electrodeposits. In addition, the hydrophobic additive, i.e., acetic acid, which suppresses the coalescence of evolved hydrogen bubbles and thereby makes the pore size controllable, proved to be detrimental to the formation of a well-defined network structure. This led to a non-uniform or discontinuous copper network. This implies that acetic acid critically retards the electrodeposition of copper.

• The Journal of Korean Institute of Electromagnetic Engineering and Science
• /
• v.17 no.8 s.111
• /
• pp.760-766
• /
• 2006

In this paper, we present a wideband patch antenna with folded and shorted structure for 5 GHz WLAN(Wireless Local Area Network). The proposed antenna used folded and shorted structure in the rectangular patch for miniaturization and wide frequency bandwidth. The antenna was designed by using 3D simulation program, HFSS(High Frequency Structure Simulator) software of the Ansoft company and the implemented antenna was measured by using HP 8720c network analyzer and far field measurement chamber. Simulation result on the return loss shows fairly good characteristic of at least 13.41dB in whole frequency range of interests, and the 10dB bandwidth is 1,523MHz which shows wide bandwidth characteristic. And the simulated maximum gain of the proposed antenna is 6.57 dBi at 5.825GHz. Measured result for the 10dB bandwidth of the implemented folded and shorted structure antenna is 1,377 MHz. Measured maximum gain of the implemented antenna is 6.87dBi at 5.775GHz. Measured results for the implemented antenna showed applicable performances for the 5 GHz WLAN.