• Journal of Broadcast Engineering
• /
• v.19 no.3
• /
• pp.316-328
• /
• 2014

In this paper, we parallelize the popular feature detection algorithms, i.e. SIFT and SURF, and its application to fast panoramic image generation on the latest embedded GPU. Parallelized algorithms are implemented using recently developed OpenCL as the embedded GPGPU software platform. We compare the implementation efficiency and speed performance of conventional OpenGL Shading Language and OpenCL. Experimental result shows that implementation on OpenCL has comparable performance with GLSL. Compared with the performance on the embedded CPU in the same application processor, the embedded GPU runs 3~4 times faster. As an example of using feature extraction, panorama image synthesis is performed on embedded GPU by applying image matching using detected features.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.48 no.4
• /
• pp.12-20
• /
• 2011

In this paper, we propose an efficient object detection and classification algorithm for video surveillance applications. Previous researches mainly concentrated either on object detection or classification using particular type of feature e.g., Scale Invariant Feature Transform (SIFT) or Speeded Up Robust Feature (SURF) etc. In this paper we propose an algorithm that mutually performs object detection and classification. We combinedly use heterogeneous types of features such as texture and color distribution from local patches to increase object detection and classification rates. We perform object detection using spatial clustering on interest points, and use Bag of Words model and Naive Bayes classifier respectively for image representation and classification. Experimental results show that our combined feature is better than the individual local descriptor in object classification rate.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.40 no.8
• /
• pp.1655-1666
• /
• 2015

In this paper, user-position estimation method is proposed by using a single camera for both indoor and outdoor environments. Conventionally, the GPS of RF-based estimation methods have been widely studied in the literature for outdoor and indoor environments, respectively. Each method is useful only for indoor or outdoor environment. In this context, this study adopts a vision-based approach which can be commonly applicable to both environments. Since the distance or position cannot be extracted from a single still image, the reference images pro-stored in image database are used to identify the current position from the single still image captured by a single camera. The reference image is tagged with its captured position. To find the reference image which is the most similar to the current image, the SURF algorithm is used for feature extraction. The outliers in extracted features are discarded by using RANSAC algorithm. The performance of the proposed method is evaluated for two buildings and their outsides for both indoor and outdoor environments, respectively.

• Journal of Broadcast Engineering
• /
• v.16 no.4
• /
• pp.614-623
• /
• 2011

We propose a real-time tracking algorithm for an augmented reality (AR) system for TV broadcasting. The tracking is initialized by detecting the object with the SURF algorithm. A multi-scale approach is used for the stable real-time camera tracking. Normalized cross correlation (NCC) is used to find the patch correspondences, to cope with the unknown and changing lighting condition. Since a zooming camera is used, the focal length should be estimated online. Experimental results show that the focal length of the camera is properly estimated with the proposed online calibration procedure.

• Journal of the Semiconductor & Display Technology
• /
• v.14 no.3
• /
• pp.67-72
• /
• 2015

This paper proposes an object tracking system based on keypoints using homography in mobile environments. The proposed system is based on markerless tracking, and there are four modules which are recognition, tracking, detecting and learning module. Recognition module detects and identifies an object to be matched on current frame correspond to the database using LSH through SURF, and then this module generates a standard object information. Tracking module tracks an object using homography information that generate by being matched on the learned object keypoints to the current object keypoints. Then update the window included the object for defining object's pose. Detecting module finds out the object based on having the best possible knowledge available among the learned objects information, when the system fails to track. The experimental results show that the proposed system is able to recognize and track objects with updating object's pose for the use of mobile platform.

• Journal of the Semiconductor & Display Technology
• /
• v.14 no.3
• /
• pp.73-77
• /
• 2015

This paper proposes an object tracking system according to reliability assessment of learning in mobile environments. The proposed system is based on markerless tracking, and there are four modules which are recognition, tracking, detecting and learning module. Recognition module detects and identifies an object to be matched on current frame correspond to the database using LSH through SURF, and then this module generates a standard object information that has the best reliability of learning. The standard object information is used for evaluating and learning the object that is successful tracking in tracking module. Detecting module finds out the object based on having the best possible knowledge available among the learned objects information, when the system fails to track. The experimental results show that the proposed system is able to recognize and track the reliable objects with reliability assessment of learning for the use of mobile platform.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.21 no.3
• /
• pp.296-301
• /
• 2011

This paper deals with the simultaneous localization and mapping system using cooperative robot. For recognizing environment, swarm robot uses the ultrasonic sensors and vision sensor. Ultrasonic sensors measure the distance information, and vision sensor recognizes the predefined landmark. we used SURF with excellent quality and fast matching in order to recognize landmark. Due to measurement error of sensors, we fusion them using particle filter for accurate localization and mapping. Finally, we show the feasibility of the proposed method through some experiments.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.36 no.6
• /
• pp.483-497
• /
• 2018

Image mosaicking, which combines several images into one image, is effective for analyzing images and important in various fields of spatial information such as a continuous image map. The crucial processes of the image mosaicking are optimal seamline determination and color correction of mosaicked images. In this study, the overlap regions were determined by SURF (Speeded Up Robust Features) for image matching. Based on the characteristics of the edges extracted by Canny filter, seamline candidates were selected from classified edges with their characteristics, and the edges were connected by using Dijkstra algorithm. In particular, anisotropic filter and image pyramid were applied to extract reliable seamlines. In addition, it was possible to determine seamlines effectively and efficiently by utilizing building and road layers from digital maps. Finally, histogram matching and seamline feathering were performed to improve visual quality of the mosaicked images.

• Journal of Internet Computing and Services
• /
• v.16 no.3
• /
• pp.21-32
• /
• 2015

Since spatio-temporal local features for video representation have become an important issue of modeless bottom-up approaches in action recognition, various methods for feature extraction and description have been proposed in many papers. In particular, BoF(bag of features) has been promised coherent recognition results. The most important part for BoF is how to represent dynamic information of actions in videos. Most of existing BoF methods consider the video as a spatio-temporal volume and describe neighboring 3D interest points as complex volumetric patches. To simplify these complex 3D methods, this paper proposes a novel method that builds BoF representation as a way to learn 2D interest points directly from video data. The basic idea of proposed method is to gather feature points not only from 2D xy spatial planes of traditional frames, but from the 2D time axis called spatio-temporal frame as well. Such spatial-temporal features are able to capture dynamic information from the action videos and are well-suited to recognize human actions without need of 3D extensions for the feature descriptors. The spatio-temporal BoF approach using SIFT and SURF feature descriptors obtains good recognition rates on a well-known actions recognition dataset. Compared with more sophisticated scheme of 3D based HoG/HoF descriptors, proposed method is easier to compute and simpler to understand.

• Proceedings of the Korean Information Science Society Conference
• /
• 2011.06c
• /
• pp.421-424
• /
• 2011

최근에 스마트폰이 발달하고 대부분의 모바일 기기에 카메라가 달리면서 카메라를 이용한 애플리케이션 또한 늘어나고 있는데 기존의 PC상에서 로고 인식등을 위해 사용되는 SURF를 이용한 이미지 매칭에는 유클리드 거리 계산을 사용하고 있다. 그러나 이 방법으로는 PC보다는 사양이 낮은 모바일 기기에 적용하기에는 기존에 사용하고 있는 방법이 인식할 이미지마다 모든 특징점을 비교하는 방법을 사용하기 때문에 연산량이 높은 편이다. 본 논문에서는 미리 인식할 이미지를 뉴럴넷에 학습시킨 뒤, 뉴럴넷을 필터링으로 사용하여 일부의 특징점만을 비교해 연산량을 줄여서 속도를 향상시키는 방법을 제안하였으며 이를 이용하여 대략 30%가량의 성능 향상이 나타난 것을 알 수 있었다.