• Electronics and Telecommunications Trends
• /
• v.34 no.6
• /
• pp.133-144
• /
• 2019

Although deep learning-based visual image recognition technology has evolved rapidly, most of the commonly used methods focus solely on recognition accuracy. However, the demand for low latency and low power consuming image recognition with an acceptable accuracy is rising for practical applications in edge devices. For example, most Internet of Things (IoT) devices have a low computing power requiring more pragmatic use of these technologies; in addition, drones or smartphones have limited battery capacity again requiring practical applications that take this into consideration. Furthermore, some people do not prefer that central servers process their private images, as is required by high performance serverbased recognition technologies. To address these demands, the object and scene recognition technologies for mobile/embedded devices that enable optimized neural networks to operate in mobile and embedded environments are gaining attention. In this report, we briefly summarize the recent trends and issues of object and scene recognition technologies for mobile and embedded devices.

• 제어로봇시스템학회:학술대회논문집
• /
• 2005.06a
• /
• pp.2211-2216
• /
• 2005

This paper describes a novel navigation algorithm using a locomotion interface with two 6-DOF parallel robotic manipulators. The suggested novel navigation system can induce user's real walking and generate realistic visual feedback during navigation, using robotic manipulators. For realistic visual feedback, the virtual environment is designed with three components; 3D object modeler for buildings and terrains, scene manager and communication manager component. The walking velocity of the user is directly translated to VR actions for navigation. Finally, the functions of the RPC interface are utilized for each interaction mode. The suggested navigation system can allow a user to explore into various virtual terrains with real walking and realistic visual feedback.

• Journal of the Korea Society of Computer and Information
• /
• v.21 no.9
• /
• pp.37-43
• /
• 2016

This paper introduces a method for video shot group detection needed for efficient management and summary of video. The proposed method detects shots based on low-level visual properties and performs temporal and spatial clustering based on visual similarity of neighboring shots. Shot groups created from temporal clustering are further clustered into small groups with respect to visual similarity. A set of representative shot frames are selected from each cluster of the smaller groups representing a scene. Shots excluded from temporal clustering are also clustered into groups from which representative shot frames are selected. A number of video clips are collected and applied to the method for accuracy of shot group detection. We achieved 91% of accuracy of the method for shot group detection. The number of representative shot frames is reduced to 1/3 of the total shot frames. The experiment also shows the inverse relationship between accuracy and compression rate.

• Journal of the Korean Institute of Landscape Architecture
• /
• v.30 no.4
• /
• pp.19-27
• /
• 2002

When we research a landscape or make a Landscape Impact Assessment, we use the image of a specific season like summer or fall. Since there are four distinct seasons, each with a different landscape, researchers need to understand viewers′visual preferences for individual seasonal landscapes. The purpose of this study is to investigate viewers′visual preferences according to seasonal change and the respondent′s age, gender and profession. In this research, the independent variable is season: suing, summer, fall, winter and snowy winter. Three landscape types used in the experiment: forest, street and agriculture. Each landscape type has two sites for reliability. The assessment media for this research are pictures featuring landscapes taken in each of the four seasons. The study used the "paired comparison" method for taking the score of visual preference. The results of this study are as follows: 1. The summer landscape has the highest visual preference score. However, spring and fall landscapes should also be considered for visual landscape evaluation. 2. The visual preference of winter landscape covered with snow is very high, but since snow is temporal and irregular, it is difficult to consider this factor for visual landscape evaluation. 3. The visual preference score of winter is the lowest of four seasons. The attractive factors of spring are flowers, summer is greenery and fall is autumnal tints. But these are not present in winter. 4. The result of visual preferences according to age groups, gender and profession have no serious differences. 5. Visual preference to scenery of 4 seasons by age group was not different from general preference and thus was concluded to have no connection with age. 6. As a result from the research of visual preference to scenery of 4 seasons by sex, women were shown to like snow-scene more than men. This study presents an indication of general preferences of seasonal landscapes. It is expected that more advanced study will proceed after this one.

• Journal of KIISE:Software and Applications
• /
• v.27 no.8
• /
• pp.864-876
• /
• 2000

This paper proposes a new scene boundary detection scheme for the MPEG System stream using MPEG Audio information and proves its usefulness by extensive experiments. A scene boundary has a characteristic that the audio as well as video information are changed rapidly. This paper first classifies this scene boundary into three cases ; Radical, Gradual, Micro Changes, with respect to the audio changes. The Radical change has a large-scale changing of decibel value and pitch value at a scene boundary, the Gradual change shows the long-time transition of decibel and pitch values from max to min or vice versa, and the Micro change displays a some change of pitch or frequency distribution without decibel changes. Upon this analysis, a new scene change detection algorithm detecting these three cases is proposed in which a progressive window with a time line is used to trace the changes in the audio information. Some experiments with various movies show that proposed algorithm could produce a high detection ratio for Radical change that is the most popular scene change in the movies, while producing a moderate detection ratio for Gradual and Micro changes. The proposed scene boundary detection scheme could be used to build a database for visual information like MPEG System stream.

• Journal of KIISE:Computing Practices and Letters
• /
• v.12 no.6
• /
• pp.358-366
• /
• 2006

This paper describes a novel navigation algorithm using a locomotion interface with two 6-DOF programmable foot platforms. When a human walks on the locomotion interface (LI), the walking motions of the human are recognized by several sensors. Then, the sensed information is used by the LI for generation of infinite surfaces for continuous walking and the virtual environments for scene update according to motions of the human walking. The suggested novel navigation system can induce user's real walking and generate realistic visual feedback during navigation. A novel navigation algorithm is suggested to allow natural navigation in virtual environments by utilizing conditions of normal gait analysis. For realistic visual feedback, the virtual environment is designed with three components; 3D object modeler for buildings and terrains, scene manager and communication manager component. From experiments, the subjects were satisfied with the reality of the suggested navigation algorithm using the locomotion interface. Therefore, the suggested navigation system can allow a user to explore into various virtual terrains with real walking and realistic visual feedback.

• Journal of Broadcast Engineering
• /
• v.17 no.1
• /
• pp.1-16
• /
• 2012

In order to improve the workflow of 3D video production, this paper analyzes the visual fatigue caused from distortions in 3D video production stage through a set of subjective visual assessment tests. To establish a set of objective indicators for subjective visual tests, various distortions in production stage are investigated to be categorized into 7 representative visual-fatigue-producing factors, and to conduct visual assessment tests for each selected category, 4 test video clips are produced by combining the extent of camera movement as well as the object(s) movement in the scene. Each produced test video is distorted to reflect each of the selected 7 visual-fatigue-producing factors, and we set 7 levels of distortion for each factor, resulting in 196 5-second-long video clips for testing. Based on these test materials and the recommendation of ITU-R BT.1438, subjective visual assessment tests are conducted by 101 applicants. The test results provide a relative importance and the tolerance limit of each visual-fatigue-producing factor, which corresponds to various distortions in 3D video production field.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.371-374
• /
• 1997

Recently, the development of computer achieves a system which is similar to the mechanics of human visual system. The 3-dimensional measurement using monocular vision system must be achieved a camera calibration. So far, the camera calibration technique required reference target in a scene. But, these methods are inefficient because they have many calculation procedures and difficulties in analysis. Therefore, this paper proposes a native method that without reference target in a scene. We use the grid type frame with different line widths. This method uses vanishing point concept that possess a rotation parameter of the camera and perspective ration that perspect each line widths into a image. We confirmed accuracy of calibration parameter estimation through experiment on the algorithm with a grid paper with different line widths.

• Proceedings of the KIEE Conference
• /
• 1998.07g
• /
• pp.2333-2335
• /
• 1998

Recently, the development of computer achieves a system which is similar to the mechanics of human visual system. The 3D measurement using monocular vision system must be achieved a camera calibration. So far, the camera calibration technique required reference target in a scene. But, these methods are inefficient because they have many calculation procedures and difficulties in analysis. Therefore, this paper proposes a native method that without reference target in a scene. We use the grid type frame with different line widths. This method uses vanishing point concept that possess a rotation parameter of the camera and perspective ration that perfect each line widths into a image. We confirmed accuracy of calibration parameter estimation through experiment on the algorithm with a grid paper with different line widths.

• Journal of the Semiconductor & Display Technology
• /
• v.16 no.2
• /
• pp.70-74
• /
• 2017

Given two stereo images of a scene, it is possible to recover a 3D understanding of the scene. This is the primary way that the human visual system estimates depth. This process is useful in applications like robotics, where depth sensors may be expensive but a pair of cameras is relatively cheap. In this work, we combined our interests to implement a graph cut algorithm for stereo correspondence, and performed evaluation against a baseline algorithm using normalized cross correlation across a variety of metrics. Experimental trials revealed that the proposed descriptor exhibited a significant improvement, compared to the other existing methods.