• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.6
• /
• pp.842-849
• /
• 2022

With the development of deep learning technology, there are many cases of using DNNs in embedded systems such as unmanned vehicles, drones, and robotics. Typically, in the case of an autonomous driving system, it is crucial to run several DNNs which have high accuracy results and large computation amount at the same time. However, running multiple DNNs simultaneously in an embedded system with relatively low performance increases the time required for the inference. This phenomenon may cause a problem of performing an abnormal function because the operation according to the inference result is not performed in time. To solve this problem, the solution proposed in this paper first reduces the computation by applying the Tucker decomposition to DNN models with big computation amount, and then, make DNN models run in parallel as much as possible in the unit of hidden layer inside the GPU. The experimental result shows that the DNN inference time decreases by up to 75.6% compared to the case before applying the proposed technique.

• The Journal of the Convergence on Culture Technology
• /
• v.7 no.4
• /
• pp.705-710
• /
• 2021

Nowadays, cutting-edge information communication technology is the genuine core technology of the fourth Industrial Revolution and is still making great progress rapidly among various technology fields. The biggest issue in ICT fields is the machine learning based Artificial Intelligence applications using big data in cloud computing environment on the basis of wireless network, and also the technology fields of autonomous control applications such as Autonomous Car or Mobile Robot. Since value of the high-tech ICT technology depends on the surrounded environmental factors and is very flexible, the precise technology valuation method is urgently needed in order to get successful technology transfer, transaction and commercialization. In this research, we analyze the characteristics of the high-tech ICT technology and the main factors in technology transfer or commercialization process, and propose the precise technology valuation method that reflects the characteristics of the ICT technology through phased analysis of the existing technology valuationmodel.

• Journal of Broadcast Engineering
• /
• v.26 no.6
• /
• pp.738-747
• /
• 2021

Today, with the improvement of deep learning technology, computer vision areas such as image classification, object detection, object segmentation, and object tracking have shown remarkable improvements. Various applications such as intelligent surveillance, robots, Internet of Things, and autonomous vehicles in combination with deep learning technology are being applied to actual industries. Accordingly, the requirement of an efficient compression method for video data is necessary for machine consumption as well as for human consumption. In this paper, we propose an object-based compression of thermal infrared images for machine vision. The input image is divided into object and background parts based on the object detection results to achieve efficient image compression and high neural network performance. The separated images are encoded in different compression ratios. The experimental result shows that the proposed method has superior compression efficiency with a maximum BD-rate value of -19.83% to the whole image compression done with VVC.

• KIPS Transactions on Software and Data Engineering
• /
• v.11 no.12
• /
• pp.509-516
• /
• 2022

Accurate pose prediction of objects in 3D space is an important visual recognition technique widely used in many applications such as scene understanding in both indoor and outdoor environments, robotic object manipulation, autonomous driving, and augmented reality. Most previous works for object pose estimation have the limitation that they require an exact 3D CAD model for each object. Unlike such previous works, this paper proposes a novel neural network model that can predict the poses of unknown objects based on only their RGB color images without the corresponding 3D CAD models. The proposed model can obtain depth maps required for unknown object pose prediction by using an adaptive depth estimator, AdaBins,. In this paper, we evaluate the usefulness and the performance of the proposed model through experiments using benchmark datasets.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• 2022.11a
• /
• pp.388-389
• /
• 2022

Recently, AI (Artificial Intelligence) has been applied to various technologies such as automatic driving, robot and smart communication. Currently, AI system is developed by software-based method using tensor flow, and GPU (Graphic Processing Unit) is employed for processing unit. However, if software-based method employing GPU is used for AI applications, there is a problem that we can not change the internal circuit of processing unit. In this method, if high-level jobs are required for AI system, we need high-performance GPU, therefore, we have to change GPU or graphic card to perform the jobs. In this work, we developed a CNN-based FPGA (Field Programmable Gate Array) chip to solve this problem.

• Journal of IKEEE
• /
• v.27 no.4
• /
• pp.494-500
• /
• 2023

Image processing is currently used in various fields. Among them, autonomous vehicles, medical image processing, and robot control require fast image processing response speeds. To fulfill this requirement, hardware design for real-time processing is being actively researched. In addition to the size of the input image, the hardware processing speed is affected by the size of the inactive video periods that separate lines and frames in the image. In this paper, we design three different scaler structures based on the type of line memories, which is closely related to the inactive video periods. The structures are designed in hardware using the Verilog standard language, and synthesized into logic circuits in a field programmable gate array environment using Xilinx Vivado 2023.1. The synthesized results are used for frame rate analysis while comparing standard image sizes that can be processed in real time.

• Journal of Drive and Control
• /
• v.20 no.4
• /
• pp.27-34
• /
• 2023

Weeding in orchards is closely associated with productivity and quality. The customary weeding process is both labor-intensive and time-consuming. To solve the problems, there is need for automation of agricultural robots and machines in the agricultural field. On the other hand, orchards have complicated working areas due to narrow spaces between trees and amorphous terrain. Therefore, it is necessary to develop customized robot technology for unmanned weeding work within the department. This study developed a path generation and path control method for unmanned weeding according to the orchard environment. For this, the width of the weeding span, the number of operations, and the width of the weeding robot were used as input parameters for the orchard environment parameters. To generate a weeding path, a weeding robot was operated remotely to obtain GNSS-based location data along the superheated center line, and a driving performance test was performed based on the generated path. From the results of orchard field tests, the RMSE in weeding period sections was measured at 0.029 m, with a maximum error of 0.15 m. In the steering period within row and steering to the next row sections, the RMSE was 0.124 m, and 0.047 m, respectively.

• Journal of the Korea Society of Computer and Information
• /
• v.29 no.8
• /
• pp.23-32
• /
• 2024

NVS (Novel View Synthesis) is a field in computer vision that reconstructs new views of a scene from a set of input views. Real-time rendering and high performance are essential for NVS technology to be effectively utilized in various applications. Recently, 3D-GS (3D Gaussian Splatting) has gained popularity due to its faster training and inference times compared to those of NeRF (Neural Radiance Fields)-based methodologies. However, since 3D-GS reconstructs a 3D (Three-Dimensional) scene by splitting and cloning (Density Control) Gaussian points, the number of Gaussian points continuously increases, causing the model to become heavier as training progresses. To address this issue, we propose two methodologies: 1) Gaussian blending, an improved density control methodology that removes unnecessary Gaussian points, and 2) a performance enhancement methodology using a depth estimation model to minimize the loss in representation caused by the blending of Gaussian points. Experiments on the Tanks and Temples Dataset show that the proposed methodologies reduce the number of Gaussian points by up to 4% while maintaining performance.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.4
• /
• pp.28-34
• /
• 2010

In this paper, we proposed an algorithm for utilizing visual information for non-contact predicting method of friction coefficient. Coefficient of friction is very important in driving on road and traversing over obstacle. Our algorithm is based on terrain classification for visual image. The proposed method, non-contacting approach, has advantage over other methods that extract material characteristic of road by sensors contacting road surface. This method is composed of learning group(experiment, grouping material) and predicting friction coefficient group(Bayesian classification prediction function). Every group include previous work of vision. Advantage of our algorithm before entering such terrain can be very useful for avoiding slippery areas. We make experiment on measurement of friction coefficient of terrain. This result is utilized real friction coefficient as prediction method. We show error between real friction coefficient and predicted friction coefficient for performance evaluation of our algorithm.

• Journal of KIISE
• /
• v.45 no.1
• /
• pp.36-44
• /
• 2018

Due to the advent of the 4th Industrial Revolution, it is imperative that we aggressively continue to develop state-of-the-art, cutting edge ICT technology relative to autonomous vehicles, intelligent robots, and so forth. Especially, systems based on convergence IT are being developed in the form of CPSs (Cyber Physical Systems) that interwork with sensors and actuators. Since conventional CPS specification only expresses behavior of one system, specification for collaboration and diversity of CPS systems with characteristics of hyper-connectivity and hyper-convergence in the 4th Industrial Revolution has been insufficiently presented. Additionally, behavioral modeling of CPSs that considers more collaborative characteristics has been unachieved in real-time application domains. This study defines the non-functional requirements that should be identified in developing embedded software for real-time constrained collaborating CPSs. These requirements are derived from ISO 25010 standard and formally specified based on state-based timed process. Defined non-functional requirements may be reused to develop the requirements for new embedded software for CPS, that may lead to quality improvement of CPS.