• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.20 no.6
• /
• pp.313-330
• /
• 2021

In an autonomous driving system equipped with a camera and lidar, depth completion techniques enable dense depth estimation. In particular, using self-supervised learning it is possible to train the depth completion network even without ground truth. In actual autonomous driving, such depth completion should have very short latency as it is the input of other algorithms. So, rather than complicate the network structure to increase the accuracy like previous studies, this paper focuses on network latency. We design a U-Net type network with RegNet encoders optimized for GPU computation. Instead, this paper presents several techniques that can increase accuracy during the process of self-supervised learning. The proposed techniques increase the robustness to unreliable lidar inputs. Also, they improve the depth quality for edge and sky regions based on the semantic information extracted in advance. Our experiments confirm that our model is very lightweight (2.42 ms at 1280x480) but resistant to noise and has qualities close to the latest studies.

• Journal of Broadcast Engineering
• /
• v.27 no.6
• /
• pp.906-913
• /
• 2022

In this paper, we propose a new deep network architecture using nearest neighbor kernel for the estimation of dense depth map from its sparse map and corresponding color information. First, we propose to decompose the depth map signal into the structure and details for easier prediction. We then propose two separate subnetworks for prediction of both structure and details using classification and regression approaches, respectively. Moreover, the nearest neighboring kernel method has been newly proposed for accurate prediction of structure signal. As a result, the proposed method showed better results than other methods quantitatively and qualitatively.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.1350-1352
• /
• 2022

본 논문에서는 희소깊이영상과 컬러영상을 이용해 조밀한 깊이영상을 추정하는 깊이 완성(depth completion)을 수행하기위해 최근접 이웃 커널을 추정하는 방식의 네트워크를 제안한다. 회귀방식의 딥러닝 네트워크는 일반적으로 값을 직접 예측하는 것보다 기본 값에 더해질 잔차를 추정하는 방식이 더욱 효율적이다. 본 논문에서는 최근접 이웃 커널을 입력영상에 적용하여 추정하고자 하는 픽셀의 인근 픽셀에서 값을 가져와 기본 값으로 사용하고, 해당 값의 잔차를 회귀방식으로 추정하는 네트워크를 설계했다. 이러한 방식으로 여러 SOTA 알고리즘 대비 좋은 성능을 나타냈고, 특히 이와 유사한 방식인 Plane-residual net 보다 높은 성능을 보여준다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2016.10a
• /
• pp.689-690
• /
• 2016

밀집된 돈방에서 사육되는 돼지의 공격적인 행동들은 돼지의 성장에 심각한 악영향을 주고, 이는 농가의 경제적 손실로 이어진다. 따라서 돈방 내의 비정상 상황들을 지속적으로 모니터링할 수 있는 IT기반의 영상 감시 시스템이 요구된다. 본 논문에서는 돼지의 행동 분석 이전에 필수적으로 선행되어야 하는 개별 돼지의 탐지를 위한 키넥트 카메라 기반의 새로운 모니터링 시스템을 제안한다. 먼저, 배경차영상 기법과 깊이 임계값을 이용하여 서있는 돼지만을 탐지한다. 둘째, 서있는 돼지들 중에서 움직임이 있는 돼지만을 관심영역으로 설정하여 탐지한다. 마지막으로, 서서 움직이는 돼지들 사이에서 발생하는 근접 문제를 깊이 정보를 이용한 등고선기법을 제안 적용하여 돼지 객체의 탐지를 완성한다. 실제 세종에 위치한 한 돈사에서 취득한 깊이 영상 정보를 이용하여 본 논문에서 제안하는 시스템의 성능을 실험적으로 검증하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.313-316
• /
• 2018

최근 미디어의 생성 및 소비 기술의 발전으로 몰입도 있는 콘텐츠에 대한 수요가 증가하고 있다. View Interpolation 기술은 두 개의 좌/우 영상을 기반으로 하여 두 영상의 중간 시점에 해당하는 영상을 생성해내는 기술이다. 먼저 Depth Hole Filling Module을 이용하여 좌/우 영상 및 그에 대응하는 깊이 지도를 입력으로 받아 깊이 지도에 존재하는 오류를 검출하고, 보정한다. 깊이 지도의 오류 보정이 완료되면, 해당 데이터를 각각 Feature Matching Module 및 Layer Dividing Module로 전달한다. Feature Matching Module은 실사 영상 내의 특징점들을 검출하고, 두 영상 내 특징점을 매칭하는 역할을 수행하며, Layer Dividing Module은 깊이 값을 기반으로 영상의 Layer를 분할한다. Feature Matching Module에서 특징점의 매칭이 완료되면, 특징점의 영상 내 좌표 및 해당 좌표에서의 깊이 값을 Distance Estimating Module로 전달한다. Distance Estimating Module은 전달받은 특징점의 좌표 및 해당 좌표에서의 깊이 값을 기반으로 전체 깊이 값에서의 이동도를 계산한다. 이와 같이 이동도의 계산 및 Layer 분할이 완료되면, 각 Layer를 이동도에 기반하여 이동시키고, 이동된 Layer들을 포개어 배치함으로써 View interpolation을 완성한다.

• Journal of Digital Contents Society
• /
• v.16 no.4
• /
• pp.645-650
• /
• 2015

In this paper, we propose the method for creating a stable stereoscopic 3D contents with the production guidelines by removing the excessive depth value and scene changes for high quality. We have developed a three-dimensional depth analysis tool for detecting the scene changes out of the production guidelines and the depth value changes excessively. The Scenes detected by depth analysis tool can be modified at the post production and it helps to make a stable stereoscopic 3D contents.

• KIPS Transactions on Computer and Communication Systems
• /
• v.5 no.10
• /
• pp.319-326
• /
• 2016

Abnormal situation caused by aggressive behavior of pigs adversely affects the growth of pigs, and comes with an economic loss in intensive pigsties. Therefore, IT-based video surveillance system is needed to monitor the abnormal situations in pigsty continuously in order to minimize the economic demage. In this paper, we propose a new Kinect camera-based monitoring system for the detection of the individual pigs. The proposed system is characterized as follows. 1) The background subtraction method and depth-threshold are used to detect only standing-pigs in the Kinect-depth image. 2) The moving-pigs are labeled as regions of interest. 3) A contour method is proposed and applied to solve the touching-pigs problem in the Kinect-depth image. The experimental results with the depth videos obtained from a pig farm located in Sejong illustrate the efficiency of the proposed method.

• Journal of the Korea Computer Industry Society
• /
• v.7 no.5
• /
• pp.455-466
• /
• 2006

In this paper, 3D objects is created from the real scene that is used by an active sensor, which gets depth and RGB information. To get the depth information, this paper uses the $Zcam^{TM}$ camera which has built-in an active sensor module. <중략> Thirdly, calibrate the detailed parameters and create 3D mesh model from the depth information, then connect the neighborhood points for the perfect 3D mesh model. Finally, the value of color image data is applied to the mesh model, then carries out mapping processing to create 3D object. Experimentally, it has shown that creating 3D objects using the data from the camera with active sensors is possible. Also, this method is easier and more useful than the using 3D range scanner.

• Cartoon and Animation Studies
• /
• s.40
• /
• pp.157-176
• /
• 2015

This research describes render pass synthetic techniques required to use for the effectiveness of them in 3D animation synthetic technology. As the render pass is divided by property and synthesized after rendering, elaborate, rapid synthesis can be achieved. In particular, occlusion pass creates a screen as if it had a soft, light shading, expressing a sense of depth and boundary softness. It is converted into 2D image through a process of pass rendering of animation projects created in 3D space, then completed in synthetic software. Namely, 3D animation realizes the completeness of work originally planned through compositing, a synthetic process in the last half. To complete in-depth image, a scene manufactured in 3D software can be sent as a synthetic program by rendering the scene by layer and property. As recently the occlusion pass can express depth notwithstanding conducting GI rendering of 3D graphic outputs, it is an important synthetic map not omitted in the post-production process. Nonetheless, for the importance of it, currently the occlusion pass leaves much to be desired for research support and books summarizing and analyzing the characteristics of properties, and the principles and usages of them. Hence, this research was aimed to summarize the principles and usages of occlusion map, and analyze differences in the results of synthesis. Furthermore, it also summarized a process designating renderers and the map utilizing the properties, and synthetic software usages. For the future, it is hoped that effective and diverse latter expression techniques will be studied beyond the limitation of graphic expression based on trends diversifying technique development.

• Korean Journal of Plant Resources
• /
• v.13 no.2
• /
• pp.147-153
• /
• 2000

Experiments were carried out to determine the effects of culturing conditions DCI(Day required for Colonization after Inoculation), CR(Contamination Rate), MD(Mycelial Density), DPI(Day required for Primordial sclerotia formation after Inoculation), yields, and degree of the foreign materials formations in Poria cocos sclerotia. The upper and bottom side of wood logs were covered with gauze, rice hull and floating leaves, vinyl, or covering cloth in order to prevent foreign material formations in p. cocos. The major results were positive in the order of rice hull, floating leaf, vinyl, gauze, and covering cloth covered and dwindled. In case of the upper and bottom sides of the wood logs covered with rice hull, DCI required 50 days more when compared with the control group(45 days) and CR was 0%, while the control group showed 2% CR. MD was excellent in case of conventional methods as against control's. DPI required 17~20 days less when compared with the control group(82 days). Yields of p. cocos in case of the upper and bottom side of wood logs covered with rice hull were 5.87kg, which is 35.7% higher than that of the control group(4.33kg), and 5.62kg in the case of upper sides cover only(increased by 29.8% compared with control), and 5.59kg in the case of bottom side cover only(increased by 29.1% compared with control), and foreign materials were none. In a separate experiment, where the upper and bottom sides of wood logs were covered with rice hull to prevent the foreign material formation, the results were as follows : Sclerotia formation status and quality of P. cocos were effective in the order of 20cm, 40cm, and 60cm of buring depth and dwindled. In 20cm of burying depth, DCI was shortened by 5 days and CR was none, while the control group showed 2% CR. MD was remarkably fine at the mulched conditions. DPI was shortened by 20 days when compared with the control(62 days).