• Journal of KIBIM
• /
• v.7 no.1
• /
• pp.45-53
• /
• 2017

Continuous research efforts have been made on acquiring location data on construction sites. As a result, GPS and RFID are increasingly employed on the site to track the location of equipment and materials. However, these systems are based on radio frequency technologies which require attaching tags on every target entity. Implementing the systems incurs time and costs for attaching/detaching/managing the tags or sensors. For this reason, efforts are currently being made to track construction entities using only cameras. Vision-based 3D tracking has been presented in a previous research work in which the location of construction manpower, vehicle, and materials were successfully tracked. However, the proposed system is still in its infancy and yet to be implemented on practical applications for two reasons. First, it does not involve entity matching across two views, and thus cannot be used for tracking multiple entities, simultaneously. Second, the use of a checker board in the camera calibration process entails a focus-related problem when the baseline is long and the target entities are located far from the cameras. This paper proposes a vision-based method to track multiple workers simultaneously. An entity matching procedure is added to acquire the matching pairs of the same entities across two views which is necessary for tracking multiple entities. Also, the proposed method simplified the calibration process by avoiding the use of a checkerboard, making it more adequate to the realistic deployment on construction sites.

• Journal of information and communication convergence engineering
• /
• v.18 no.2
• /
• pp.75-81
• /
• 2020

Approximately 1.3 million people die from traffic accidents each year, and smartphone usage while driving is one of the main causes of such accidents. Therefore, detection of smartphone usage by drivers has become an important part of distracted driving detection. Previous studies have used single camera-based methods to collect the driver images. However, smartphone usage detection by employing a single camera can be unsuccessful if the driver occludes the phone. In this paper, we present a driver smartphone usage detection system that uses multiple cameras to collect driver images from different perspectives, and then processes these images with ensemble convolutional neural networks. The ensemble method comprises three individual convolutional neural networks with a simple voting system. Each network provides a distinct image perspective and the voting mechanism selects the final classification. Experimental results verified that the proposed method avoided the limitations observed in single camera-based methods, and achieved 98.96% accuracy on our dataset.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.193-198
• /
• 2009

Generally a TV broadcast video of ball sports is composed from those of multiple cameras strategically mounted around a stadium under the supervision of a master director. The director decides which camera the current view should be from and how the camera work should be. In this paper, such a decision rule is based on the 3D location of ball which is the result of multi-view tracking. While current TV sports broadcast are accompanied with professional cameramen and expensive equipments, our system requires few video cameras and no cameraman. The resulted videos were stable and informative enough to convey the flow of a match.

• Journal of Korea Multimedia Society
• /
• v.24 no.4
• /
• pp.538-549
• /
• 2021

In this paper, we introduce a technique for 360-degree panoramic video streaming with multiple virtual cameras in real-time. The proposed technique consists of generating 360-degree panoramic video data by ORB feature point detection, texture transformation, panoramic video data compression, and RTSP-based video streaming transmission. Especially, the generating process of 360-degree panoramic video data and texture transformation are accelerated by CUDA for complex processing such as camera calibration, stitching, blending, encoding. Our experiment evaluated the frames per second (fps) of the transmitted 360-degree panoramic video. Experimental results verified that our technique takes at least 30fps at 4K output resolution, which indicates that it can both generates and transmits 360-degree panoramic video data in real time.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2020.07a
• /
• pp.75-76
• /
• 2020

본 논문에서는 스마트 폰 등의 영상수신장치들이 원격으로 접속하여 다양한 인터페이스를 갖는 영상취득 장치들 중 하나를 선택하여 제어하면서 해당 영상을 수신할 수 있는 영상송신시스템을 제안한다. 제안하는 영상송신시스템은 우선적으로 야간의 물리적 보안을 위한 영상취득용 초저도 카메라를 위해 고정된 4개의 컴포지트 및 USB CAM 활용을 위한 인터페이스들 및 이를 위한 해당 기능들, 영상수신장치들이 무선으로 접속하여 채널을 확보하기 위한 Soft-AP 기능, 접속된 영상수신장치들에게 해당 영상을 전송하며 이렇게 연결된 다수의 장치를 제어하는 등의 전반적인 관리를 위한 기능 등으로 구성되는데, 이러한 기능들의 소프트웨어적 구현에 비중을 두어 기술한다.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.8 no.6
• /
• pp.899-905
• /
• 2013

In this paper, we focus on how to calibrate multi-cameras easily and how to efficiently detect quad-copters with small-numbered particles. Each particle is a six dimensional vector that is composed of 3D position and 3D orientation of a quad-copter in the space. Due to curse of dimensionality, that leads to explosive computational costs with a large amount of high-dimensioned particles. To detect efficiently, we need to put more particles in very promising spaces and few particles in other spaces. Though computational cost is lowered by minimizing particles, in order to track a quad-copter with multiple cameras in real-time, multiple images from the cameras should be synchronized and analyzed. Therefore, lots of the computations still need to be done. Because of this, GPGPU(General-Purpose computing on Graphics Processing Units) is implemented for parallel computing. This method has been successfully tested and gives accurate results in practical situations.

• Smart Media Journal
• /
• v.3 no.4
• /
• pp.22-27
• /
• 2014

Three-dimensional image reconstruction allows us to represent real objects in the virtual space and observe the objects at arbitrary view points. This technique can be used in various application areas such as education, culture, and art. In this paper, we propose an implementation method of the high-quality three-dimensional object using multiple Kinect cameras released from Microsoft. First, We acquire color and depth images from triple Kinect cameras; Kinect cameras are placed in front of the object as a convergence form. Because original depth image includes some areas where have no depth values, we employ joint bilateral filter to refine these areas. In addition to the depth image problem, there is an color mismatch problem in color images of multiview system. In order to solve it, we exploit an color correction method using three-dimensional geometry. Through the experimental results, we found that three-dimensional object which is used the proposed method is more naturally represented than the original three-dimensional object in terms of the color and shape.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.1808-1813
• /
• 2004

Recognition of human motions using their 2-D images has various applications. An eigenspace method is employed in this paper for representing and recognizing human motions. An eigenspace is created from the images taken by multiple cameras that surround a human in motion. Image streams obtained from the cameras compose the same number of curved lines in the eigenspace and they are used for recognizing a human motion in a video image. Performance of the proposed technique is shown experimentally.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09a
• /
• pp.628-631
• /
• 2003

In this paper, we present real-time, accurate face region detection and tracking technique for an intelligent surveillance system. It is very important to obtain the high-resolution images, which enables accurate identification of an object-of-interest. Conventional surveillance or security systems, however, usually provide poor image quality because they use one or more fixed cameras and keep recording scenes without any cine. We implemented a real-time surveillance system that tracks a moving person using four pan-tilt-zoom (PTZ) cameras. While tracking, the region-of-interest (ROI) can be obtained by using a low-pass filter and background subtraction. Color information in the ROI is updated to extract features for optimal tracking and zooming. The experiment with real human faces showed highly acceptable results in the sense of both accuracy and computational efficiency.

• International journal of advanced smart convergence
• /
• v.10 no.2
• /
• pp.45-52
• /
• 2021

The surveillance camera market has developed and plays an important role in the field of video surveillance. However, in recent years, the identification of areas requiring surveillance has been limited by reflective light in the surveillance camera market. Cameras using polarization filters are being developed to reduce reflective light and facilitate identification. Programs are required to automatically adjust polarization filters. In this paper, we proposed an optimal angle extraction method from surveillance cameras using polarization filters through histogram analysis. First of all, transformed to grayscale to analyze the specifications of frames in multiple polarized angle images, reducing computational throughput. Then we generated and analyzed a histogram of the corresponding frame to extract the angle when the highlights are the fewest. Experiments with 0˚ and 90˚ showed high performance in extracting optimal angles. At this point, it is hoped this technology would be used for surveillance cameras in place like beach with a lot of reflective light.