• Journal of the Institute of Convergence Signal Processing
• /
• v.6 no.1
• /
• pp.1-7
• /
• 2005

This paper describes the system for detecting vehicles in the rear and rear-side that access between sidewalk and bus stopped to city road at day by computer vision-based method. This system informs appearance of vehicles to bus driver and passenger for the safety of a bus passenger getting off. The camera mounted on the top portion of the bus exit door gets the rear and rear-side image of the bus whenever a bus stops at the stop. The system sets search area between bus and sidewalk from this image and detects a vehicle by using change of image and sobel filtering in this area. From a central point of the vehicle detected, we can find out the distance, speed and direction by its location, width and length. It alarms the driver and passengers when it's judged that dangerous situation for the passenger getting off happens. This experiment results in a detection rate more than 87% in driving by bus on the road.

• Journal of Broadcast Engineering
• /
• v.24 no.6
• /
• pp.1064-1071
• /
• 2019

Multiview stereo matching algorithm is used to reconstruct 3D shape from a set of 2D images. Conventional multiview stereo algorithms have been implemented on high-performance hardware due to the heavy complexity that contains a large number of calculations in each step. However, as the performance of mobile graphics processors has recently increased rapidly, complex computer vision algorithms can now be implemented on mobile devices like a smartphone and an embedded board. In this paper we parallelize an multiview stereo algorithm using OpenCL on mobile GPU and provide various optimization techniques on the embedded hardware with limited resource.

• Journal of Korea Game Society
• /
• v.10 no.2
• /
• pp.49-56
• /
• 2010

In this paper, we propose a control method of virtual hand by the recognition of a user's hand in the virtual reality game environment. We display virtual hand on the game screen after getting the information of the user's hand movement and the direction thru input images by camera. We can utilize the movement of a user's hand as an input interface for virtual hand to select and move the object. As a hand recognition method based on the vision technology, the proposed method transforms input image from RGB color space to HSV color space, then segments the hand area using double threshold of H, S value and connected component analysis. Next, The center of gravity of the hand area can be calculated by 0 and 1 moment implementation of the segmented area. Since the center of gravity is positioned onto the center of the hand, the further apart pixels from the center of the gravity among the pixels in the segmented image can be recognized as fingertips. Finally, the axis of the hand is obtained as the vector of the center of gravity and the fingertips. In order to increase recognition stability and performance the method using a history buffer and a bounding box is also shown. The experiments on various input images show that our hand recognition method provides high level of accuracy and relatively fast stable results.

• The Journal of The Korea Institute of Intelligent Transport Systems
• /
• v.20 no.5
• /
• pp.186-201
• /
• 2021

As autonomous driving technology advances, the accuracy of the vehicle position is important for recognizing the environments around driving. Map-matching localization techniques based on high definition (HD) maps have been studied to improve localization accuracy. Because conventional map-matching techniques estimate the vehicle position based on an HD map reference dataset representing the center of the lane, the estimated position does not reflect the deviation of the lateral distance within the lane. Therefore, this paper proposes a localization system based on the reference lateral position dataset extracted using image processing and HD maps. Image processing extracts the driving lane number using inverse perspective mapping, multi-lane detection, and yellow central lane detection. The lane departure method estimates the lateral distance within the lane. To collect the lateral position reference dataset, this approach involves two processes: (i) the link and lane node is extracted based on the lane number obtained from image processing and position from GNSS/INS, and (ii) the lateral position is matched with the extracted link and lane node. Finally, the vehicle position is estimated by matching the GNSS/INS local trajectory and the reference lateral position dataset. The performance of the proposed method was evaluated by experiments carried out on a highway environment. It was confirmed that the proposed method improves accuracy by about 1.0m compared to GNSS / INS, and improves accuracy by about 0.04m~0.21m (7~30%) for each section when compared with the existing lane-level map matching method.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2004.10a
• /
• pp.454-457
• /
• 2004

AGV for FMS must be detected an obstacle. Therefore, many studies have been advanced, and recently, the ultra sonic sensor is used for this. However, the new method has to be developed because the ultra-sonic-sensor has many problems as a noise in factory, an directional error and detection of the obstacle size. So, we study the fast stereo vision system that can give more information to obstacles for intelligent AGV system. For this, the simulated AGV system was made with two CCD cameras in front to get the stereo images, and the threshold process by color information (intensity and chromaticity) and structure stereo matching method were constructed.

• Transactions of the Korean Society of Machine Tool Engineers
• /
• v.16 no.1
• /
• pp.9-18
• /
• 2007

There have been continuous efforts for automation of joint tracking system. This automation process is mainly used to do in root pass of gas metal arc welding in the field of heavy industry and shipbuilding etc. For automation, it is important using of vision sensor. Welding robot with vision sensor is used for weld seam tracking on welding fabrication. Recently, it is used to on post-weld inspection for weld quality evaluation. For real time seam tracking, it is very important role in vision process technique. Vision process is included in filtering and thinning, segmentation processing, feature extraction and recognition. In this paper, it has shown performance comparison results of seam tracking for real time root pass on gas metal arc welding. It can be concluded better segment splitting method than iterative averaging technique in the performance results of seam tracking.

• Proceedings of the KAIS Fall Conference
• /
• 2005.05a
• /
• pp.227-230
• /
• 2005

최근 컴퓨터 비전 및 영상처리 분야에서 일련의 동영상으로부터 움직이는 객체를 검출하고 추적하는 시스템에 대한 관심이 점차 높아지고 있다. 본 연구에서는 원격 제어에 의해 상하좌우 이동이 가능한 장치에 고정된 카메라를 이용하여 움직이는 단일 이동 객체를 추적할 수 있는 시스템을 구현하였다. 추적 시스템은 영상 획득 및 전처리 단계, 차영상 기법에 기반한 움직임 추정 단계와 추정된 중심점 좌표를 기준으로 카메라를 제어하는 추적 단계로 구성되어 있다. 구현된 이동 객체의 실시간 추적 시스템의 성능을 평가하기 위해 사람이 보통의 속도로 실내에서 이동하는 환경에서 실시간 추적 실험을 실시하였다. 그 결과 객체를 잃어버리지 않고 정상적으로 추적할 수 있었다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2007.10c
• /
• pp.424-427
• /
• 2007

스테레오 영상으로부터 3차원 구조를 복원하기 위해서는 깊이맵에 해당하는 시차맵을 생성해야 한다. 시차맵 생성을 위해서는 정합비용을 계산하고, 집성한 후에 시차를 계산하는 절차로 이루어진다. 본 논문에서는 스테레오 영상으로부터 빠르고 안정된 시차맵을 생성하기 위해서 후처리 과정으로 각 스캔라인에 대해서 분산을 이용하여 세그멘테이션을 한 후에 세그멘테이션 별로 평균을 내어 객체간의 구분을 명확히 한다. 조밀 시차맵을 생성하기 위해서는 시차 계산에 실패한 화소들에 대해서도 시차를 계산해야 하는데 본 논문에서는 간단하게 인접 화소의 값을 복사하는 방법으로 홀을 메우는 방법을 제안한다. 실제 환경에서의 다양한 스테레오 영상에 대한 실험 결과들은 제안된 시차맵 생성과 홀을 메우는 방법이 기존의 시차맵 생성 기법만큼 빠르고 기존의 방법보다 좀더 안정적이고 다양한 컴퓨터 비전 시스템응용에 적용될 수 있음을 보여준다.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2004.05b
• /
• pp.149-153
• /
• 2004

A Impulse denosing filter design of image for the faster processing time and system compatibility using DSP processor is presented on this paper The system hardware is composed of the stand-alone board with 32 bits DSP processor and vision board for image data acquisition with NTSC CCD camera, and the host computer controls them. The denoising method uses the adaptive median filter. The experiment result is that the system leads to denosing effect as 90% and PSNR 22㏈

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.113-116
• /
• 2021

일반적으로 딥러닝 기반의 객체 검출(Object Detection)기법은 합성곱 신경망(Convolutional Neural Network, CNN)을 통해 입력된 영상의 특징(Feature)을 추출하여 이를 통해 객체 검출을 수행한다. 최근 자연어 처리 분야에서 획기적인 성능을 보인 트랜스포머(Transformer)가 영상 분류, 객체 검출과 같은 컴퓨터 비전 작업을 수행하는데 있어 경쟁력이 있음이 드러나고 있다. 본 논문에서는 YOLOv4-CSP의 CSP 블록을 개선한 one-stage 방식의 객체 검출 네트워크를 제안한다. 개선된 CSP 블록은 트랜스포머(Transformer)의 멀티 헤드 어텐션(Multi-Head Attention)과 CSP 형태의 공간 피라미드 풀링(Spatial Pyramid Pooling, SPP) 연산을 기반으로 네트워크의 Backbone과 Neck에서의 feature 학습을 돕는다. 본 실험은 MSCOCO test-dev2017 데이터 셋으로 평가하였으며 제안하는 네트워크는 YOLOv4-CSP의 경량화 모델인 YOLOv4s-mish에 대하여 평균 정밀도(Average Precision, AP)기준 2.7% 향상된 검출 정확도를 보인다.