• Journal of IKEEE
• /
• v.26 no.2
• /
• pp.160-168
• /
• 2022

In this paper, we propose a 3D mesh reconstruction method from a single image using deep learning and a sphere shape transformation method. The proposed method has the following originality that is different from the existing method. First, the position of the vertex of the sphere is modified to be very similar to the 3D point cloud of an object through a deep learning network, unlike the existing method of building edges or faces by connecting nearby points. Because 3D point cloud is used, less memory is required and faster operation is possible because only addition operation is performed between offset value at the vertices of the sphere. Second, the 3D mesh is reconstructed by covering the surface information of the sphere on the modified vertices. Even when the distance between the points of the 3D point cloud created by correcting the position of the vertices of the sphere is not constant, it already has the face information of the sphere called face information of the sphere, which indicates whether the points are connected or not, thereby preventing simplification or loss of expression. can do. In order to evaluate the objective reliability of the proposed method, the experiment was conducted in the same way as in the comparative papers using the ShapeNet dataset, which is an open standard dataset. As a result, the IoU value of the method proposed in this paper was 0.581, and the chamfer distance value was It was calculated as 0.212. The higher the IoU value and the lower the chamfer distance value, the better the results. Therefore, the efficiency of the 3D mesh reconstruction was demonstrated compared to the methods published in other papers.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2003.06a
• /
• pp.1283-1287
• /
• 2003

Reverse engineering technology refers to the process that creates a CAD model of an existing part using measuring devices. Recently, non-contact scanning devices have become more accurate and the speed of data acquisition has increased drastically. However, they generate thousands of points per second and various types of point data. Therefore. it becomes a important to handle the huge amount and various types of point data to generate a surface model efficiently. This paper proposes a new triangular mesh generation method using 3D grids. The geometric information of a part can be obtained from point cloud data by estimating normal values of the points. In our research, the non-uniform 3D grids are generated first for feature based data reduction based on the geometric information. Then, triangulation is performed with the reduced point data. The grid structure is efficiently used not only for neighbor point search that can speed up the mesh generation process but also for getting surface connectivity information to result in same topology surface with the point data. Through this integrated approach, it is possible to create surface models from scanned point data efficiently.

• KIPS Transactions on Software and Data Engineering
• /
• v.12 no.5
• /
• pp.229-236
• /
• 2023

Since 3D printed medical implant parts usually have surface defects, it is necessary to inspect the surface after manufacturing. In order to automate the surface inspection, it is effective to 3D scan the implant and reconstruct it as a scan model such as a point cloud. When constructing a scan model, the characteristics of the shape and material of the implant must be considered because it has characteristics different from those of general 3D printed parts. In this paper, we present a method to reconstruct the 3D scan model of a 3D printed metal bone-plate that is one kind of medical implant parts. Multiple partial scan data are produced by multi-view 3D scan, and then, we reconstruct a scan model by alignment and merging of partial data. We also present the process of the scan model reconstruction through experiments.

• Journal of Korea Water Resources Association
• /
• v.54 no.3
• /
• pp.181-190
• /
• 2021

In this study, a large-scale levee breach experiment from lateral overflow was conducted to verify the effect of the new reinforcement method applied to the levee's surface. The new method could prevent levee failure and minimize damage caused by overflow in rivers. The levee was designed at the height of 2.5 m, a length of 12 m, and a slope of 1:2. A new material mixed with biopolymer powder, water, weathered granite, and loess in an appropriate ratio was sprayed on the levee body's surface at a thickness of about 5 cm, and vegetation recruitment was also monitored. At the Andong River Experiment Center, a flow (4 ㎥/s) was introduced from the upstream of the A3 channel to induce the lateral overflow. The change of lateral overflow was measured using an acoustic doppler current profiler in the upstream and downstream. Additionally, cameras and drones were used to analyze the process of the levee breach. Also, a new method using 3D point cloud for calculating the surface loss rate of the levee over time was suggested to evaluate the performance of the levee reinforcement method. It was compared to existing method based on image analysis and the result was reasonable. The proposed 3D point cloud methodology could be a solution for evaluating the performance of levee reinforcement methods.

• Journal of IKEEE
• /
• v.28 no.3
• /
• pp.375-381
• /
• 2024

In this paper, we propose a lightweight object classification system using a LiDAR sensor for delivery service robots. The 3D point cloud data is encoded into a 2D pseudo image using a Pillar Feature Network (PFN), and then passed through a lightweight classification network designed based on Depthwise Separable Convolutional Neural Networks (DS-CNN). The implementation results show that the designed classification network has 9.08K parameters and 3.49M Multiply-Accumulate (MAC) operations, while supporting a classification accuracy of 94.94%.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.05a
• /
• pp.482-484
• /
• 2019

Learning and analyzing 3D point clouds with deep networks is challenging due to the limited and irregularity of the data. In this paper, we present a data-driven point cloud augmentation technique. The key idea is to learn multilevel features per point and to reconstruct to a similar point set. Our network is applied to a projection loss function that encourages the predicted points to remain on the geometric shapes with a particular target. We conduct various experiments using ShapeNet part data to evaluate our method and demonstrate its possibility. Results show that our generated points have a similar shape and are located closer to the object.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.33-36
• /
• 2021

In this paper, we propose a new method for improving compression efficiency of immersive video using reinforcement learning. Immersive video means a video that a user can directly experience, such as 3DOF+ videos and Point Cloud videos. It has a vast amount of information due to their characteristics. Therefore, lots of compression methods for immersive video are being studied, and generally, a method, which projects an 3D image into 2D image, is used. However, in this process, a region where information does not exist is created, and it can decrease the compression efficiency. To solve this problem, we propose the reinforcement learning-based filling method with considering the characteristics of images. Experimental results show that the performance is better than the conventional padding method.

• Journal of the Korean Society of Industry Convergence
• /
• v.24 no.6_2
• /
• pp.753-762
• /
• 2021

Google provides mapping services using satellite imagery, this is widely used for the study. Since about 20 years ago, research and business using drones have been expanding. Pix4D is widely used to create 3D information models using drones. This study compared the distance error by comparing the result of the road construction site with the DSM data of Google Earth and Pix4 D. Through this, we tried to understand the reliability of the result of distance measurement in Google Earth. A DTM result of 3.08 cm/pixel was obtained as a result of matching with 49666 key points for each image. The length and altitude of Pix4D and Google Earth were measured and compared using the obtained PCD. As a result, the average error of the distance based on the data of Pix4D was measured to be 0.68 m, confirming that the error was relatively small. As a result of measuring the altitude of Google Earth and Pix4D and comparing them, it was confirmed that the maximum error was 83.214m, which was measured using satellite images, but the error was quite large and there was inaccuracy. Through this, it was confirmed that there are difficulties in analyzing and acquiring data at road construction sites using Google Earth, and the result was obtained that point cloud data using drones is necessary.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.22 no.11
• /
• pp.1505-1510
• /
• 2018

Product modeling technology for graphics is rapidly developing. And 3D data application and usability are increasing.modeling of product design is a very important factor in constructing. 3D modeling in product design takes a lot of production time. Recently, the reverse design method is very useful because of application of 3D data and shortening of production time. In this study, first, 3D point cloud and mesh data are generated using photographs based on image data. The second is to modify the design and the third is to make the prototype with the 3D printer. This product design and production process suggests the utilization and possibility of image data, the shortening of 3D modeling production time and efficient processes. Also, the product design process proposes a model of a new product development system to adapt to the production environment.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.3
• /
• pp.143-151
• /
• 2009

In this paper, a new extraction algorithm for artificial structure in 3D cloud points of terrestrial LIDAR is described, considering that various obstacles in terrestrial LIDAR make it difficult to apply conventional algorithms which are designed for air-born LIDAR data. Firstly we use the R, G, B color information from the terrestrial LIDAR data to discriminate among the massive 3D cloud points. Hough transform is then applied to estimate the straight lines that correspond to the target structure. Finally, the structure is extracted by comparing the distance between the estimated line and 3D cloud points. The proposed algorithm is efficient in the sense that it requires the user interaction only when the reference colors are obtained. Computer simulation shows the performance to be quite satisfactory.