Earticle

DVR 및 NVR을 이용한 디지털 저장매체를 영상감시 시스템에서 활용하게 되면서 영상처리 모듈의 개발은 영상 보안 시장의 필수적인 요소이다. 특히 네트워크 카메라의 등장은 기존 아날로그 방식의 CCTV를 대체하면서 영 상처리 모듈 개발의 필요성을 더욱 부각시키고 있다. 본 논문에서는 움직임 감지 기법을 이용한 영상 감시 서버를 설 계 및 구축하고 서버에서 처리되는 영상처리 결과를 실시간으로 모바일 디바이스에서 확인 가능한 영상감시 시스템 을 개발하였다. 영상처리를 위해 리눅스 기반의 서버에 오픈소스 OpenCV를 활용한 영상처리 모듈을 개발하였고, 네 트워크 카메라로부터 전송되는 실시간 비디오 데이터를 저장 및 가공하여 안드로이드기반 모바일 기기에서 검색 가 능한 영상감시 시스템을 구축하였다.

It is essential to support the image and video analysis technology such as motion detection since the DVR and NVR storage were adopted in the real time visual surveillance system. Especially the network camera would be popular as a video input device. The traditional CCTV that supports analog video data get be replaced by the network camera. In this paper, we present the design and implementation of video surveillance system that provides the real time motion detection by the video storage server. The mobile application also has been implemented in order to provides the retrieval functionality of image analysis results. We develop the video analysis server with open source library OpenCV and implement the daemon process for video input processing and real-time image analysis in our video surveillance system.

This study aims to present a environmental monitoring tool which is using remote sensing technique to improve the understanding ability of beach litter condition. In order to grasp occurrence time of beach litter and its pollution area, we did the real-time monitoring with network camera. When confirm that there is a lot of litter occurred within the scope of camera, we use unmanned aerial vehicle to monitoring contaminated status of total beach. The area monitored was Heungnam Beach on Geoje Island, which is located in the south of Korea. During the monitoring period, the maximum beach litter was produced on 28th May 2013 (29.9 m2/580 pieces). In the future, the proposed monitoring method can be used in association with remote detection methods to determine the appropriate time for collecting litter and improve the effectiveness of beach litter management policies.

We focus on the issue of learning the topology of the non-overlapping multi-camera network, which includes recovering the nodes (entry and exit zones), transition time distribution and links. Firstly, the nodes associated with each camera view are identified using clustering method. Then, transition time distribution is modeled as a Gaussian distribution and is computed by accumulated cross correlation and Gaussian fitting. Finally, the mutual information is used to refine the possible links and the topology is recovered. Experimental results on simulated data and real scene demonstrate the effectiveness of the proposed method.

The calibration between a camera and a two-dimensional laser scanner (2DLS) is an essential step in the object detecting system. Many algorithms with linear model have been proposed. But these tend to solve intrinsic and extrinsic calibration parameters separately and are influenced seriously by the poor initial data, which leads to unstable and inaccurate results. Hence, a new nonlinear model based on the Back Propagation neural network trained by the Levenberg-Marquardt algorithm (LM-BP) is presented for calibration in this paper. Before the calibration, the original laser data is fitted linearly to avoid the ranging error and is optimized by an angular increment to reduce the step-angular error. Then, the calibration network with 4 inputs composed of the lasers points’ coordinates and constant 1, and 2 outputs are obtained, expected values of which are the coordinates of corresponding points in the image coordinates. The sum of square of errors between the network outputs and expected values is taken to adjust the modifications of the weights and thresholds with the Levenberg-Marquardt method to optimize the calibration model. Finally, compared with related researches, experimental results show that the accuracy of calibration between camera and 2DLS is significantly improved, and the detecting system is more suitable for actual measurement situations.

In this paper, we propose the intelligent unmanned anti-theft system using network cameras. To do this, we use two Internet Protocol (IP) cameras, which are installed at the inside and the outside of the restricted area. First of all, the external camera is used to monitor a potential intruder entering the restricted area in real time. To detect the potential intruder from sequential images obtained by external IP camera, we propose the robust algorithm combining the background modeling, the hybrid silhouette, the convex summation, the morphology and the smoothing. Once the intruder is detected by the external camera, the internal camera with pan-tilt mechanism is then activated and starts tracking the intruder and monitoring the prescribed valuables registered with some feature points in the database. To track the intruder, we propose the tracking algorithm using the background elimination and the histogram. And then, we completes the proposed anti-theft system by using the Scale Invariant Feature Transform (SIFT) algorithm in order to monitor whether the intruder steals the prescribed valuables or not. Finally, we have shown the effectiveness and the applicability of the proposed method through experiments.

Zoom tracking involves the automatic adjustment of the focus motor in response to the zoom motor movements for the purpose of keeping an object of interest in focus, and is typically achieved by moving the zoom and focus motors in a zoom lens module so as to follow the so-called "trace curve", which shows the in-focus motor positions versus the zoom motor positions for a specific object distance. Thus, one can simply implement zoom tracking by following the most closest trace curve after all the trace curve data are stored in memory. However, this approach is often prohibitive in practical implementation because of its large memory requirement. Many other zoom tracking methods such as GZT, AZT and etc. have been proposed to avoid large memory requirement but with a deteriorated performance. In this paper, we propose a new zoom tracking method called 'Approximate Feedback Zoom Tracking method (AFZT)' on DM36x-based IP network camera, which does not need large memory by approximating nearby trace curves, but generates better zoom tracking accuracy than GZT or AZT by utilizing focus value as feedback information. Experiments through real implementation shows the proposed zoom tracking method improves the tracking performance and works in real-time.

In this paper, we developed a network camera for CMOS camera module inspection. The design, implementation details including embedded linux porting and CPLD logics, and performance of network camera are described. The network camera consists of SoC(S3C4530A), CPLD and CMOS image sensor. In order to image data of CMOS image sensor we designed capture logics on CPLD by using VHDL program. Embedded Linux such as uClinux is performed on the network camera to utilize development environment and TCP/IP protocol specification. The application is based on socket communication between GUI on PC and Embedded Linux based network camera. When JPEG compression is applied, the transmission speed was improved enough for this system to be used for an alternative of expensive CCTV or remote monitoring system in a power plant and uninhabited places.

PC camera-based instant messenger have problems of video image break/delay and noise. The reason is caused by a composed problem that PC Camera is belong to the connected PC and operating in the PC. This research suggests an instant messenger that uses network camera connected to LAN instead of PC camera. Therefore, the instant messenger has a component that the camera is separated from PC. This research designs sequence diagram, video calls flow chart, and graphic user interface of the instant messenger. The instant messenger based on network camera provides high quality video calls service and makes it possible to call from a special remote site with real-time video image.

인터넷 웹 기반의 Home Securiy, ITS(Intelligent Traffic System), 관광산업, 생산현장 등 여러 분야에서 네트워크 카메라를 이용한 영상정보 시스템이 각광을 받고 있으며, 이에 따른 네트워크 카메라의 수요가 급속하게 증가하고 있다. 또한 이를 제어하기 위해서 비디오 서버가 복잡해짐에 따라 많은 비용이 드는 문제를 가지고 있다. 따라서 본 논문에서는 카메라 수의 증가에 따른 비디오 서버의 복잡성과 비용 문제를 해결하고자 다중 채널을 통해 입력되는 카메라의 영상 정보를 1채널 멀티플랙스 스위칭 처리를 하고 또한 영상 데이터를 자동으로 스위칭하는 시스템을 구현 하였다.

Internet of the Web-based Home Securiy, ITS (Intelligent Traffic System), the tourism industry, production field, etc In many fields, using a network camera imaging system has been spotlighted and Accordingly the demand for network cameras is growing rapidly. in order to control it according to the video server complex has a costly problem. In this paper, according to an increasing number of cameras cost and complexity of the video server problems to solve information from video cameras through multi-channel input single-channel multiplex and the fact that switching is handled and Also, the system automatically switches the image data is implemented.

본 논문에서는 특정 지역을 연속해서 촬영하는 고정된 카메라 영상에서 사람들의 움직임을 검출하고 움직임을 분석하여 정량화 하는 방법을 제안한다. 먼저 배경 영상을 획득하기 위하여 일정 시간동안의 입력 영상을 누적하고 평균값으로 정규화 한다. 그리고 영상을 계속 누적하여 배경 영상을 실시간으로 갱신한다. 다음으로 획득된 배경 영상과 현재 영상에 대하여 차영상과 이진화를 수행하고 팽창 연산과 연결 성분 분석으로 잡영을 제거한다. 그리고 잡영이 제거된 영상에서 원근감 효과를 고려하는 가중치를 적용하여 움직임이 있는 객체를 클러스터링 하는 수정된 ART2 클러스터링 방법을 제안한다. 마지막으로 클러스터링 결과 정보를 이용하여 움직임을 정량화 한다. 제안하는 방법을 실내 환경에 설치된 네트워크 카메라로부터 영상을 획득하여 실험한 결과, 영상의 원근감 효과에 따라 군집 크기가 차이남에도 강인하게 분석할 수 있음을 확인하였다.

This paper presents a method for moving objects detection, analysis and expression how much move as numerical value from the image which is captured by a network camera. To perform this method, we process few kinds of pre-processing to remove noise that are getting background image, difference image, binarization and so on. And to consider perspective effect, we propose modified ART2 algorithm. Finally, we express the result of ATR2 clustering as numerical value. This method is robust to size of object which is changed by perspective effect.

자동 백색 보정은 디지털 비디오 프로세싱에서 매우 중요한 기능이며, IP 네트워크 카메라의 경우에도 필수적인 영상 전처리 과정이다. 본 논문은 TI 사의 비디오 프로세서인 DM36x을 이용하여 제작된 HD IP 네트워크 카메라에서의 실시간으로 동작할 수 있는 효과적인 비디오 자동 백색 보정 알고리즘을 구현한 결과를 기술한다. 구현한 자동 백색 보정 알고리즘은 매우 효과적인 것으로 알려진 그레이 컬러 포인트 기반 알고리즘을 DM36x의 AWB 엔진이 제공하는 기능을 이용하여 실시간으로 동작될 수 있도록 고쳐서 구현하였다.

보안이 중요시 되는 공간에서 지능형 감시 시스템의 필요성이 점차 중요시 되고 있다. 본 논문에서는 embedded Linux 기반의 Mobile Robot에 Network Camera를 탑재 하여 침입자를 추적할 수 있는 시스템 구현에 목적을 두고 있다. Network Camera부터 Wireless Lan을 이용하여 서버로 영상을 전송하고, 서버에서 블록매칭 알고리즘을 이용하여 침입자의 이동경로를 파악하며 침입자에 대한 방향 정보를 전송하여 침입자를 추적한다. 로봇이 침입자를 추적함에 따라 침입자의 유효 영상을 얻는다. 본 논문에 의해서 구현된 시스템은 다른 감시 시스템과 연동하여 지능형 감시 시스템으로서 신뢰성을 더할 수 있다.

ONVIF is becoming a de factor standard specification for supporting interoperability among network video products, which also supports a specification for video analytics service. A smart IP network camera is an IP network supporting video analytics. In this paper, we present our efforts in integrating ONVIF Video Analytics Service into our currently developing smart IP network camera(SS IPNC; Soongsil Smart IP Network Camera). SSIPNC supports object detection, tracking, classification, and event detection with proprietary configuration protocol and meta data formats. SSIPNC is based on TI' IPNC ONVIF implementation which supports ONVI Core specification, and several ONVIF services such as device service, imaging service and media service, but not video analytics service.

This paper deals with an application of neural network to camera calibration with wide angle lens and 2-D range finding. Wide angle lens has an advantage of having wide view angles for mobile environment recognition ans robot eye in hand system. But, it has severe radial distortion. Multilayer neural network is used for the calibration of the camera considering lens distortion, and is trained it by error back-propagation method. MLP can map between camera image plane and plane the made by structured light. In experiments, Calibration of camers was executed with calibration chart which was printed by using laser printer with 300 d.p.i. resolution. High distortion lens, COSMICAR 4.2mm, was used to see whether the neural network could effectively calibrate camera distortion. 2-D range of several objects well be measured with laser range finding system composed of camera, frame grabber and laser structured light. The performance of 3-D range finding system was evaluated through experiments and analysis of the results.

This paper proposes a novel indoor positioning system (IPS) that uses a calibrated camera sensor network and dense 3D map information. The proposed IPS information is obtained by generating a bird's-eye image from multiple camera images; thus, our proposed IPS can provide accurate position information when objects (e.g., the mobile robot or pedestrians) are detected from multiple camera views. We evaluate the proposed IPS in a real environment with moving objects in a wireless camera sensor network. The results demonstrate that the proposed IPS can provide accurate position information for moving objects. This can improve the localization performance for mobile robot operation.

When a camera is employed for 3D sensing, accurate camera calibration is vital as it is a prerequisite for the subsequent steps of the sensing process. Camera calibration is usually performed by complex mathematical modeling and geometric analysis. On the other contrary, data learning using an artificial neural network can establish a transformation relation between the 3D space and the 2D camera image without explicit camera modeling. However, a neural network requires a large amount of accurate data for its learning. A significantly large amount of time and work using a precise system setup is needed to collect extensive data accurately in practice. In this study, we propose a two-step neural calibration method that is effective when only a small amount of learning data is available. In the first step, the camera projection transformation matrix is determined using the limited available data. In the second step, the transformation matrix is used for generating a large amount of synthetic data, and the neural network is trained using the generated data. Results of simulation study have shown that the proposed method as valid and effective.

기존의 멀티 카메라 시스템은 넓은 영역을 커버하거나 이동 중인 물체를 트래킹 하기 위한 목적으로 주로 사용되어 왔다. 하지만 이러한 시스템은 하나의 카메라가 커버하는 영상이 가려지면 정보를 잃게 되는 단점이 있다. 멀티 카메라 시스템은 하나의 영역을 여러 카메라가 커버하도록 하여 이런 단점을 극복할 수 있다. 또한 다양한 시점의 카메라에서 수집되는 영상의 경우, 영상에 따라 담고 있는 정보가 다르므로 여러 카메라의 입력 정보를 함께 활용하여 보다 많은 정보를 얻을 수도 있다. 본 논문은 이런 장점을 활용하여 멀티 카메라 환경에서의 이벤트 인식 문제를 다룬다. 이를 위해 사무실 환경에 8대의 카메라를 설치하였으며, 시나리오에 따라 영상을 수집하였다. 수집된 영상은 전문가에 의해 어노테이션 된 후 인식 모델의 학습에 사용되며, 학습된 베이지안 네트워크 모델의 구조와 파라미터를 도메인 지식에 기반해서 수정하여 최종 이벤트 인식 모델을 설계하였다. 실험 결과 제안하는 이벤트 인식 모델의 인식률은 평균 87.0%로 Naive Bayes보다 우수한 성능을 보임을 확인하였다.

DVS 카메라는 인간의 눈을 모델링하여 만들어져서 화면의 변화에 반응하여 Address - Event - Representation 데이터를 생성하고 이 데이터는 jAER Viwer를 통해 확인할 수 있다. 이렇게 생성된 DVS 카메라의 데이터를 Spiking Neural Network의 입력으로 주기 위해 GPU를 이용한 Spiking Neural Network 시뮬레이터인 GPUSNN과 jAER 사이에 인터페이스가 필요하다. 이 인터페이스를 이용하면 GPUSNN을 통해 비전 알고리즘을 빠르고 효과적으로 Spiking Neural Network 시뮬레이션을 할 수 있을 것이다.

DVS 카메라는 인간의 눈을 모델링하여 만들어져서 화면의 변화에 반응하여 Address - Event - Representation 데이터를 생성하고 이 데이터는 jAER Viwer를 통해 확인할 수 있다. 이렇게 생성된 DVS 카메라의 데이터를 Spiking Neural Network의 입력으로 주기 위해 GPU를 이용한 Spiking Neural Network 시뮬레이터인 GPUSNN과 jAER 사이에 인터페이스가 필요하다. 이 인터페이스를 이용하면 GPUSNN을 통해 비전 알고리즘을 빠르고 효과적으로 Spiking Neural Network 시뮬레이션을 할 수 있을 것이다.