• The Journal of the Convergence on Culture Technology
• /
• v.10 no.3
• /
• pp.909-914
• /
• 2024

This study suggests a quantitative evaluation of transfer learning, which is widely used in various AI fields, including image recognition for robot vision. Quantitative and qualitative analyses of results applying transfer learning are presented, but transfer learning itself is not discussed. Therefore, this study proposes a quantitative evaluation of transfer learning itself based on MNIST, a handwritten digit database. For the reference network, the change in recognition accuracy according to the depth of the transfer learning frozen layer and the ratio of transfer learning data and pre-training data is tracked. It is observed that when freezing up to the first layer and the ratio of transfer learning data is more than 3%, the recognition accuracy of more than 90% can be stably maintained. The transfer learning quantitative evaluation method of this study can be used to implement transfer learning optimized according to the network structure and type of data in the future, and will expand the scope of the use of robot vision and image analysis AI in various environments.

• Journal of KIISE:Software and Applications
• /
• v.37 no.4
• /
• pp.273-281
• /
• 2010

In this paper, we propose an automatic segmentation method of the endo- and epicardial boundary by using ray-casting profile based on intensity distribution and gradient information in CT images. First, endo-cardial boundary points are detected by using adaptive thresholding and seeded region growing. To include papillary muscles inside the boundary, the endo-cardial boundary points are refined by using ray-casting based profile. Second, epi-cardial boundary points which have both a myocardial intensity value and a maximum gradient are detected by using ray-casting based adaptive gradient profile. Finally, to preserve an elliptical or circular shape, the endo- and epi-cardial boundary points are refined by using elliptical interpolation and B-spline curve fitting. Then, curvature-based contour fitting is performed to overcome problems associated with heterogeneity of the myocardium intensity and lack of clear delineation between myocardium and adjacent anatomic structures. To evaluate our method, we performed visual inspection, accuracy and processing time. For accuracy evaluation, average distance difference and overalpping region ratio between automatic segmentation and manual segmentation are calculated. Experimental results show that the average distnace difference was $0.56{\pm}0.24mm$. The overlapping region ratio was $82{\pm}4.2%$ on average. In all experimental datasets, the whole process of our method was finished within 1 second.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2001.10a
• /
• pp.727-730
• /
• 2001

멀티미디어 환경의 급속한 발전에 의해 영상처리 기술은 인간의 인체와 관련하여 얼굴인식, 제스처 인식에 관한 응용과 더불어 스포츠 관련분야로 깊숙히 정착하고 있다. 그러나 입력영상으로부터 움직이고 있는 선수들의 동작을 추출 및 추적하는 일은 컴퓨터비전 연구의 난 문제 중의 하나로 알려져 있다. 이러한 축구경기의 TV 중계에 있어서 하이라이트 장면의 자동추출(자동색인)은 그 경기의 가장 집약적인 표현이며, 축구경기 전체를 한 눈에 파악할 수 있도록 해주는 요약(summary)이자 intensive actions이고 경기의 진수이다. 따라서 축구경기와 같이 비교적 기 시간(대체로 1시간 30분) 동안 다수의 선수(양 팀 합해서 22명)들이 서로 복잡하게 뒤얽히면서 진행하는 경기의 하이라이트 장면을 효과적으로 포착하여 표현해 줄 수 있다면 TV를 통해서 경기를 관람하는 시청자들에게는 경기의 진행상황을 한 눈에 효과적으로 파악할 수 있게 해주어 흥미진진한 경기관람을 할 수 있게 해주고, 경기의 진행자들(감독, 코치, 선수 등)에게는 고차원적이고 과학적인 정보를 효과적으로 제공함으로써 한층 진보된 경기기법을 개발하고 과학적인 경기전략을 세울 수 있게 해준다. 본 논문은 이상과 같이 팀 스포츠(Team Spots)의 일종인 축구경기 하이라이트 장면의 자동색인을 위해 뉴럴네트워크 기법을 이용하여 그룹 포메이션(Group Formation) 중의 공격패턴 자동분류 기법을 개발하고 이를 검증하였다. 본 연구에서는 축구경기장 내의 빈번하게 변화하는 장면들을 자동으로 분할하여 대표 프레임을 선정하고, 대표 프레임 상에서 선수들의 위치정보와 공의 위치정보 등을 기초로 하여 경기 중에 이루어지는 선수들의 그룹 포메이션을 추적하여 그룹행동(group behavior)을 분석하고, 뉴럴네트워크의 BP(Back-Propagation) 알고리즘을 사용하여 축구경기 공격패턴을 자동으로 인식 및 분류함으로써 축구경기 하이라이트 장면의 자동추출을 위한 기반을 마련하였다. 본 연구의 실험에는 '98 프랑스 월드컵 축구경기의 다양한 공격패턴에 대한 비디오 영상에서 각각 좌측공격 60개, 우측공격 74개, 중앙공격 72개, 코너킥 39개, 프리킥 52개의 총 297개의 데이터를 추출하여 사용하였다. 실험과는 좌측공격 91.7%, 우측공격 100%, 중앙공격 87.5%, 코너킥 97.4%, 프리킥 75%로서 매우 양호한 인식율을 보였다.

• Journal of Korean Society of Transportation
• /
• v.27 no.5
• /
• pp.97-111
• /
• 2009

Image-based traffic information collection systems have entered widespread adoption and use in many countries since these systems are not only capable of replacing existing loop-based detectors which have limitations in management and administration, but are also capable of providing and managing a wide variety of traffic related information. In addition, these systems are expanding rapidly in terms of purpose and scope of use. Currently, the utilization of image processing technology in the field of traffic accident management is limited to installing surveillance cameras on locations where traffic accidents are expected to occur and digitalizing of recorded data. Accurately recording the sequence of situations around a traffic accident in a signal intersection and then objectively and clearly analyzing how such accident occurred is more urgent and important than anything else in resolving a traffic accident. Therefore, in this research, we intend to present a technology capable of overcoming problems in which advanced existing technologies exhibited limitations in handling real-time due to large data capacity such as object separation of vehicles and tracking, which pose difficulties due to environmental diversities and changes at a signal intersection with complex traffic situations, as pointed out by many past researches while presenting and implementing an active and environmentally adaptive methodology capable of effectively reducing false detection situations which frequently occur even with the Gaussian complex model analytical method which has been considered the best among well-known environmental obstacle reduction methods. To prove that the technology developed by this research has performance advantage over existing automatic traffic accident recording systems, a test was performed by entering image data from an actually operating crossroad online in real-time. The test results were compared with the performance of other existing technologies.

• KEPCO Journal on Electric Power and Energy
• /
• v.5 no.3
• /
• pp.149-156
• /
• 2019

In the field of maintenance of power transmission lines, drones have been used for their patrol and inspection by KEPCO since 2017. This drone technology was originally developed by KEPCO Research Institute, and now workers from four regional offices of KEPCO have directly applied this technology to the drone patrol and inspection tasks. In the drone inspection system, a drone with an optical zooming camera and a thermal camera can fly automatically along the transmission lines by the ground control system developed by KEPCO Research Institute, but its camera gimbal has been remotely controlled by a field worker. Especially the drone patrol and inspection has been mainly applied for the transmission lines in the inaccessible areas such as regions with river-crossings, sea-crossings and mountains. There are often communication disruptions between the drone and its remote controller in such extreme fields of mountain areas with many barriers. This problem may cause the camera gimbal be out of control, even though the inspection drone flies along the flight path well. In addition, interference with the reception of real-time transmitted videos makes the field worker unable to operate it. To solve these problems, we have developed the auto-tracking camera gimbal system with deep learning method. The camera gimbal can track the transmission line automatically, even when the transmitted video on a remote controller is intermittently unavailable. To show the effectiveness of our camera gimbal system, its field test results will be presented in this paper.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.4
• /
• pp.97-107
• /
• 2009

In image-based gait recognition systems, physical factors, such as the camera angle and the lens distortion, and environmental factors such as illumination determines the performance of recognition. In this paper we present a robust gait recognition method by compensating various types of image distortions. The proposed method is compared with existing gait recognition algorithm with consideration of both physical and environmental distortion factors in the input image. More specifically, we first present an efficient compensation algorithm of image distortion by using the projective transform, and test the feasibility of the proposed algorithm by comparing the recognition performances with and without the compensation process. Proposed method gives universal gait data which is invariant to both distance and environment. Gained data improved gait recognition rate about 41.5% in indoor image and about 55.5% in outdoor image. Proposed method can be used effectively in database(DB) construction, searching and tracking of specific objects.

• Journal of Intelligence and Information Systems
• /
• v.12 no.1
• /
• pp.1-15
• /
• 2006

In Korea, a resident registration card has various personal information such as a present address, a resident registration number, a face picture and a fingerprint. A plastic-type resident card currently used is easy to forge or alter and tricks of forgery grow to be high-degree as time goes on. So, whether a resident card is forged or not is difficult to judge by only an examination with the naked eye. This paper proposed an automatic recognition method of a resident card which recognizes a resident registration number by using a refined ART2-based RBF network newly proposed and authenticates a face picture by a template image matching method. The proposed method, first, extracts areas including a resident registration number and the date of issue from a resident card image by applying Sobel masking, median filtering and horizontal smearing operations to the image in turn. To improve the extraction of individual codes from extracted areas, the original image is binarized by using a high-frequency passing filter and CDM masking is applied to the binaried image fur making image information of individual codes better. Lastly, individual codes, which are targets of recognition, are extracted by applying 4-directional contour tracking algorithm to extracted areas in the binarized image. And this paper proposed a refined ART2-based RBF network to recognize individual codes, which applies ART2 as the loaming structure of the middle layer and dynamicaly adjusts a teaming rate in the teaming of the middle and the output layers by using a fuzzy control method to improve the performance of teaming. Also, for the precise judgement of forgey of a resident card, the proposed method supports a face authentication by using a face template database and a template image matching method. For performance evaluation of the proposed method, this paper maked metamorphoses of an original image of resident card such as a forgey of face picture, an addition of noise, variations of contrast variations of intensity and image blurring, and applied these images with original images to experiments. The results of experiment showed that the proposed method is excellent in the recognition of individual codes and the face authentication fur the automatic recognition of a resident card.

• Journal of Korea Game Society
• /
• v.6 no.2
• /
• pp.3-12
• /
• 2006

Recently, Many 3D augmented reality games that provide strengthened immersive have appeared in the 3D game environment. In this article, we describe a barehanded interaction method based on human hand gestures for augmented reality games. First, feature points are extracted from input video streams. Point features are tracked and motion of moving objects are computed. The shape of the motion trajectories are used to determine whether the motion is intended gestures. A long smooth trajectory toward one of virtual objects or menus is classified as an intended gesture and the corresponding action is invoked. To prove the validity of the proposed method, we implemented two simple augmented reality applications: a gesture-based music player and a virtual basketball game. In the music player, several menu icons are displayed on the top of the screen and an user can activate a menu by hand gestures. In the virtual basketball game, a virtual ball is bouncing in a virtual cube space and the real video stream is shown in the background. An user can hit the virtual ball with his hand gestures. From the experiments for three untrained users, it is shown that the accuracy of menu activation according to the intended gestures is 94% for normal speed gestures and 84% for fast and abrupt gestures.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.10 no.4
• /
• pp.477-482
• /
• 2015

In this paper, an android platform with Unity engine was proposed for the effective 3D presentation of IoT sensor's position to IoT(Internet of Things) users. This android platform based home-gateway was designed with the cooperation of 3D unity engine for 3D texture according to MovieTexture simultaneously. As a result, the 3D presentation technology of IoT sensors was described with Unity based 3D modeling. In this proposed smart gateway, the 3D position was presented with the received RSSI(Received Signal Strength Indicator) and angle of IoT sensors. This 3D Unity launcher can be used for the 3D position, monitoring, and the life managing of IoT sensors with the beacon and 3 dimensional cube-style after the 3D conversion of 2D.

• Proceedings of the KSR Conference
• /
• 2008.11b
• /
• pp.30-36
• /
• 2008

The Spatial Image contents of Geomorphology 3-D environment is focused by the requirement and importance in the fields such as, national land development plan, telecommunication facility management, railway construction, general construction engineering, Ubiquitous city development, safety and disaster prevention engineering. The currently used DEM system using contour lines, which embodies geographic information based on the 2-D digital maps and facility information has limitation in implementation in reproducing the 3-D spatial city. Moreover, this method often neglects the altitude of the rail way infrastructure which has narrow width and long length. There it is needed to apply laser measurement technique in the spatial target object to obtain accuracy. Currently, the LiDAR data which combines the laser measurement skill and GPS has been introduced to obtain high resolution accuracy in the altitude measurement. In this paper, we tested of the railway facilities using laser surveying system, then we propose data a generation of spatial images for the optimal manage and synthesis of railway facility system in our 3-D spatial terrain information. For this object, LiDAR based height data transformed to DEM, and the realtime unification of the vector via digital image mapping and raster via exactness evaluation is transformed to make it possible to trace the model of generated 3-dimensional railway model with long distance for 3D tract model generation. As the results, We confirmed the solutions of varieties application for railway facilities management using 3-D spatial image contents.