• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2017.11a
• /
• pp.13-15
• /
• 2017

최근 카메라로부터 획득한 영상을 이용하여 지능적으로 상황을 인지하고 판단하고 결정하는 알고리즘의 연구개발이 활발하게 이루어지고 있다. 예를 들어 영상 내의 객체를 검출하는 알고리즘, 영상 내 화재와 같은 위험 상황을 알려주는 알고리즘 등이 있다. 본 논문에서는 SuBSENSE 라고 하는 영상 내 객체를 검출하는 알고리즘에 대해서 분석하고자 한다. SuBSENSE 는 background subtraction 기반으로 동작하는 객체 알고리즘으로서 다양한 상황에도 강건하게 객체를 추출하기위한 몇 가지 과정들이 존재한다. 본 논문에서는 SuBSENSE 알고리즘 구조 분석 및 해당 구조에서 동작하는 파라미터들의 역할에 대해 살펴보고자 한다.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.13 no.6
• /
• pp.1263-1268
• /
• 2018

In this paper, a video analysis is proposed to implement video surveillance system with deep learning object detection and probabilistic data association filter for tracking multiple objects, and suggests its implementation using GPU. The proposed video analysis technique involves object detection and object tracking sequentially. The deep learning network architecture uses ResNet for object detection and applies probabilistic data association filter for multiple objects tracking. The proposed video analysis technique can be used to detect intruders illegally trespassing any restricted area or to count the number of people entering a specified area. As a results of simulations and experiments, 48 channels of videos can be analyzed at a speed of about 27 fps and real-time video analysis is possible through RTSP protocol.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.8
• /
• pp.1117-1123
• /
• 2021

Object recognition is to identify the location and class of a specific object by analyzing the given image when a specific image is input. One of the fields in which object recognition technology is actively applied in recent years is autonomous vehicles, and this paper describes the trend of image-based object recognition artificial intelligence technology in autonomous vehicles. The image-based object detection algorithm has recently been narrowed down to two methods (a single-step detection method and a two-step detection method), and we will analyze and organize them around this. The advantages and disadvantages of the two detection methods are analyzed and presented, and the YOLO/SSD algorithm belonging to the single-step detection method and the R-CNN/Faster R-CNN algorithm belonging to the two-step detection method are analyzed and described. This will allow the algorithms suitable for each object recognition application required for autonomous driving to be selectively selected and R&D.

• Journal of Korea Multimedia Society
• /
• v.13 no.8
• /
• pp.1128-1141
• /
• 2010

This paper proposes a method of moving object detection in crowded scene using clustered trajectory. Unlike previous appearance based approaches, the proposed method employes motion information only to isolate moving objects. In the proposed method, feature points are extracted from input frames first and then feature tracking is followed to create feature trajectories. Based on an assumption that feature points originated from the same objects shows similar motion as the object moves, the proposed method detects moving objects by clustering trajectories of similar motions. For this purpose an energy function based on spatial proximity, motion coherence, and temporal continuity is defined to measure the similarity between two trajectories and the clustering is achieved by minimizing the energy function in CRFs (conditional random fields). Compared to previous methods, which are unable to separate falsely merged trajectories during the clustering process, the proposed method is able to rearrange the falsely merged trajectories during iteration because the clustering is solved my energy minimization in CRFs. Experiment results with three different crowded scenes show about 94% detection rate with 7% false alarm rate.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.10 no.2
• /
• pp.183-188
• /
• 2015

In this paper, it is proposed to detect and track pedestrian and analyse tracking performance with nighttime CCTV video. The detection is performed by a cascade classifier with Haar-like feature trained with Adaboost algorithm. Tracking pedestrian is performed by a particle filter. As results of experiments, it is introduced that efficient number of particles and the distributions are applied to track pedestrian at the night-time. Performance of detection and tracking is verified with nighttime CCTV video that is obtained at alleys etc.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2002.12a
• /
• pp.301-304
• /
• 2002

그레이 스케일 영상의 객체 분할시 경계면의 모호함이 발생하여 정확하게 객체를 분할.검출 할 수 없으며 또한 영상 레벨에 따라 결과에 많은 영향을 미치게 된다 본 논문에서는 그 경계 부분을 정확하게 분할 . 검출하는 방법으로 적응적 형상학을 웨이브렛 알고리즘에 적용한후 최적의 영상을 찾는 알고리즘을 구현하였다. 본 논문은 입력 영상의 임계값에 따른 적응적 형태학을 이용하여 영상의 경계면을 레벨 업 시킨 후, 이를 웨이브렛에 적용하여 최적의 에지를 검출하였다. 또한, 기존의 영상 에지 검출알고리즘인 Sobel 에지 검출과 다른 웨이브렛 기저 계수를 적용한 에지 검출 방법과 비교하고, 제안된 알고리즘이 기존의 다른 에지 검출보다 우수함을 확인하였다. 특히 에지와 에지의 부분이 가까울 때 정확한 에지를 검출하였으며, 완만한 곡선을 가지고 있는 부분에서 더 우수한 결과 에지를 얻을 수 있음을 확인하였다.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.1
• /
• pp.9-16
• /
• 2010

Moving object detection is to detect foreground object different from background scene in a new incoming image frame and is an essential ingredient process in some image processing applications such as intelligent visual surveillance, HCI, object-based video compression and etc. Most of previous object detection algorithms are still computationally heavy so that it is difficult to develop real-time multi-channel moving object detection in a workstation or even one-channel real-time moving object detection in an embedded system using them. Foreground mask correction necessary for a more precise object detection is usually accomplished using morphological operations like opening and closing. Morphological operations are not computationally cheap and moreover, they are difficult to be rendered to run simultaneously with the subsequent connected component labeling routine since they need quite different type of processing from what the connected component labeling does. In this paper, we first devise a fast and precise foreground mask correction algorithm, "Neighbor Foreground Pixel Propagation (NFPP)" which utilizes neighbor pixel checking employed in the connected component labeling. Next, we propose a novel moving object detection method based on the devised foreground mask correction algorithm, NFPP where the connected component labeling routine can be executed simultaneously with the foreground mask correction. Through experiments, it is verified that the proposed moving object detection method shows more precise object detection and more than 4 times faster processing speed for a image frame and videos in the given the experiments than the previous moving object detection method using morphological operations.

• The Journal of the Korea Contents Association
• /
• v.11 no.4
• /
• pp.215-224
• /
• 2011

There have been studying many researches to detect human body and to track one with increasing interest on human and computer interaction. In this paper, we propose the algorithm that automatically extracts joints, linked points of human body, using the ratio of human body under single camera and tracks object. The proposed method gets the difference images of the grayscale images and ones of the hue images between input image and background image. Then the proposed method composes the results, splits background and foreground, and extracts objects. Also we standardize the ratio of human body using face' length and the measurement of human body and automatically extract joints of the object using the ratio and the corner points of the silhouette of object. After then, we tract the joints' movement using block-matching algorithm. The proposed method is applied to test video to be acquired through a camera and the result shows that the proposed method automatically extracts joints and effectively tracks the detected joints.

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.359-361
• /
• 1999

내용 기반 영상 검색을 위해 본 연구에서는 Vector Quantization을 이용하여 영상을 구성하는 주요 객체들의 특징 추출 방법을 제안한다. 내용 기반 검색 시스템에서 사용되는 영상의 주요 특징들은 색상, 질감, 형태 및 영상을 구성하고 있는 객체들의 공간적 위치 등이 사용된다. 이러한 특징들 중에서 어떤 특징들을 사용하고 또 어떤 방식으로 결합하느냐에 따라 혹은 영상의 특성을 잘 나타낼 수 있는 주요 특징을 어떻게 추출, 표현하느냐에 따라 검색 성능에 큰 영향을 미친다. 이 중 본 논문에서는 일반적인 색상, 질감 특징 추출방법과 더불어 Vector Quantization 알고리즘을 이용하여 정지 영상을 구성하고 있는 객체들의 대표 색상과 질감 특징을 빠르게 추출하고 이를 내용 기반 검색에 이용함으로써 객체의 위치, 회전 및 크기 변화에 무관한 검색을 가능케 했다. 연구의 실험 결과 VQ를 이용함으로써 대표특징치 추출시간을 줄일 수 있었고 검색시 색상과 질감 특징의 가중치를 각각 0.5, 0.5로 주는 것이 가장 높은 검출율을 보였으며 제안된 방식에 의해 '사람' 영상의 경우 0.9의 검출율을 보였다.

• Journal of Korea Multimedia Society
• /
• v.13 no.1
• /
• pp.47-57
• /
• 2010

The object detection is essential for identifying objects, location information, and user context-aware in the image. In this paper, we propose a robust object detection system. The System linearly transforms learning data obtained from the background images to Principal components. It organizes the Eigen-background with the selected Principal components which are able to discriminate between foreground and background. The Fuzzy-C-means (FCM) carries out clustering for images with inputs from the Eigen-background information and classifies them into objects and backgrounds. It used various patterns of backgrounds as learning data in order to implement a system applicable even to the changing environments, Our system was able to effectively detect partial movements of a human body, as well as to discriminate between objects and backgrounds removing noises and shadows without anyone frame image for fixed background.