Jong-Seung Park

Interactive 3D reconstruction from multiple images: A primitive-based approach

This paper proposes a fast and stable image-based modeling method which generates 3D models with high-quality face textures in a semi-automatic way. The modeler guides untrained users to quickly obtain 3D model data via several steps of simple user interface operations using predefined 3D primitives. The proposed method contains an iterative non-linear error minimization technique in the model estimation step with an error function based on finite line segments instead of infinite lines. The error corresponds to the difference between the observed structure and the predicted structure from current model parameters. Experimental results on real images validate the robustness and the accuracy of the algorithm.

Shape-Based image retrieval using invariant features

In this paper we propose an accurate shape-based image retrieval method which uses both types of moment invariants and Fourier descriptors. The method first excludes irrelevant images of different appearances using the region-based moment invariant features and then, using Fourier descriptors, the method increases the retrieval effectiveness substantially. The retrieval method was tested on our shape databases and the search accuracy was compared with those of Fourier descriptors and moment invariants. The hybrid method of using both boundary-based Fourier descriptor and region-based moment invariants provides much better performance than other similarity method.

Increasing camera pose estimation accuracy using multiple markers

If the geometry of a marker is known and camera parameters are available, it is possible to recover a camera pose. The transformation between a camera and a marker is defined relative to the local coordinate system of the marker. This paper proposes a real-time camera tracking method using multiple markers while the camera is allowed to move freely in a 3D space. We utilize multiple markers to improve the accuracy of the pose estimation. We also present a coordinate registration algorithm to obtain a global optimal camera pose from local transformations of multiple markers. For the registration, a reference marker is automatically chosen among multiple markers and the global camera pose is computed using all local transforms weighted by marker detection confidence rates. Experimental results show that the proposed method provides more accurate camera poses than those from other methods.

The Indirect Keyboard Control System by Using the Gaze Tracing Based on Haar Classifier in OpenCV

This paper describes on the indirect interface system in which general users assign computer instructions just through the gaze tracing, without mouse or keyboard. We use the web camera to replace the computer-input system. The face region and the eye region was extracted based on Haar classifier implemented on Open source Computer Vision Library (OpenCV). It controls mouse-moving by automatically affecting the position where eyesight focuses on, and simulates mouse-click by affecting blinking action. We use the virtual keyboard displayed on monitor to simulate the keyboard entry. As a result, more than 95% of the tracing accuracy was achieved when the size of a single key is larger than 25 pixels.

Vision-based real-time camera match moving using a known marker

We propose a fast and stable camera match-moving method aimed for real-time augmented reality applications. A known marker is used for fast and robust detection of marker points and for stable camera pose estimation. From the detected image points of a known marker on a single frame, we estimate the camera position and translation parameters. The entire pose estimation process is linear and initial estimates are not required. The most suitable reference point for the pose estimation is automatically chosen by investigating the depth variation. A planar marker is also allowed, since our linear method handles rank deficient cases. As an application of the proposed method, we implement a video augmentation system that replaces the marker in image frames with a virtual graphical object during the marker tracking. Experimental results show that the proposed camera tracking method is robust and fast enough for interactive video-based applications.

Vision-Based real-time camera matchmoving with a known marker

A primary requirement for practical augmented reality systems is a method of accurate and reliable camera tracking. In this paper, we propose a fast and stable camera matchmoving method aimed for real-time augmented reality application. A known marker is used for the fast detection and tracking of feature points. From the feature tracking of one of three different types of markers on a single frame, we estimate the camera position and translation parameters. The entire pose estimation process is linear and initial estimates are not required. As an application of the proposed method, we implemented a video augmentation system that replaces the marker in the image frames with a virtual 3D graphical object during the marker tracking. Experimental results showed that the proposed camera tracking method is robust and fast enough to interactive video-based applications.

Identification of scene locations from geotagged images

Due to geotagging capabilities of consumer cameras, it has become easy to capture the exact geometric location where a picture is taken. However, the location is not the whereabouts of the scene taken by the photographer but the whereabouts of the photographer himself. To determine the actual location of an object seen in a photo some sophisticated and tiresome steps are required on a special camera rig, which are generally not available in common digital cameras. This article proposes a novel method to determine the geometric location corresponding to a specific image pixel. A new technique of stereo triangulation is introduced to compute the relative depth of a pixel position. Geographical metadata embedded in images are utilized to convert relative depths to absolute coordinates. When a geographic database is available we can also infer the semantically meaningful description of a scene object from where the specified pixel is projected onto the photo. Experimental results demonstrate the effectiveness of the proposed approach in accurately identifying actual locations.

AR-Room: a rapid prototyping framework for augmented reality applications

This paper presents a novel software framework called AR-Room for fast prototyping of a variety of augmented reality applications. AR-Room consists of a lot of deployable components for core augmented reality technologies, modules for hardware abstraction, and an authoring toolkit for the rapid content design. On the AR-Room, application developers are only required to describe their content scenarios together with a configuration of software components. A content scenario is represented by a set of event-action pairs. Four major procedures in an augmented reality application are an image analyzer, an interaction handler, a rendering engine and an image synthesizer. According to the provided scenarios the designated components cooperatively provide real-time analysis and synthesis of input video frames. Several augmented reality applications are implemented on the AR-Room to show how the framework can be efficiently used for the fast prototyping of applications.

Vision-Based bare-hand gesture interface for interactive augmented reality applications

This paper presents a barehanded interaction method for augmented reality games based on human hand gestures. Point features are tracked from input video frames and the motion of moving objects is computed. The moving patterns of the motion trajectories are used to determine whether the motion is an intended gesture. A smooth trajectory toward one of virtual objects or menus is classified as an intended gesture and the corresponding action is invoked. To prove the validity of the proposed method, we implemented two simple augmented reality applications: a gesture-based music player and a virtual basketball game. The experiments for three untrained users indicate that the accuracy of menu activation according to the intended gestures is 94% for normal speed gestures and 84% for fast and abrupt gestures.

Realtime control for motion creation of 3d avatars

In this paper, we are proposing a new mechanism for controlling 3D (three dimensional) avatars to create user-designed peculiar motions of avatars in real-time using general interfaces, such as a mouse, a keyboard, or a joystick. The main idea is based on the new way of interactive control that is the combined usage of keyboard and mouse simultaneously. In order to generate natural human motions of avatars, we adopted the center line concept of art drawing and some influencing physics algorithms which developed intensively in the field of biped humanoid robot research. We demonstrate that user-designed motions of avatar can be created in real-time using proposed interaction method with keyboard and mouse. Also, we show that a rich set of peculiar behaviors can be generated from a ready-made motion with motion capture data or created and stored in our system. Note that the generated peculiar motions can be more natural if we appropriately apply our center line concept and physics algorithms.