Published papers

This paper proposes a new system for 3D object sharing with multiple clients using networks. Our system aims to support group communication across devices over the same 3D object and puts the annotations with augmented reality automatically. In order to keep putting the annotations during the interaction, we apply the vision-based 3D object tracking algorithm which estimates 3D position and pose of the object just by capturing the object with a single RGB camera. By employing the tracking algorithm robust to viewpoint changes and occlusion, we develop the 3D object sharing system with a desktop as a server and a laptop and a smart phone as clients. We conducted the experiments to show the tracking algorithm is tolerant of interaction and can be applied to various objects with texture.

This paper introduces an interface that helps understand the correspondence between the patient and medical images. Surgeons determine the extent of resection by using tomographic images such as MRI (Magnetic Resonance Imaging) data . However, understanding the relationship between the patient and tomographic images is di ffi cult. This study aims to visualize the correspond en ce more intuitive ly . In this paper, we propose an interactive visualizer for medical images based on the relative position and orientation of the handheld device and the patient. We conducted an experiment to verify the performances of the proposed method and several other methods. In the experiment, t he proposed method showed the minimum error.

We present a medical Augmented Reality (AR) edutainment system for bone anatomy learning. This learning environment, called AR bone puzzle, is a metaphor for bone anatomy learning with AR visualization and intuitive interaction. AR bone puzzle uses its user's body as a puzzle frame and computer generated virtual bones as puzzle pieces. Users learn bone anatomy by assembling the virtual bone pieces on their body. Key features of this system are 3D AR visualization and intuitive gesture based user interaction.

An efficient system that upsamples depth map captured by Microsoft Kinect while jointly reducing the effect of noise is presented. The upsampling is carried by detecting and exploiting the piecewise locally planar structures of the downsampled depth map, based on corresponding high-resolution RGB image. The amount of noise is reduced by accumulating the downsampled data simultaneously. By benefiting from massively parallel computing capability of modern commodity GPUs, the system is able to maintain high frame rate. Our system is observed to produce the upsampled depth map that is very close to the original depth map both visually and mathematically.

This paper proposes a novel view independent vehicle make and model recognition method (VMMR). Our system identifies the make and model from the variety of viewpoints while the conventional methods for VMMR work only for the fixed frontal or rear images. In addition, it needs only the 2D images not CAD data for database. To solve the alignment issue, our method uses SIFT, that has scale and rotation invariance. For the view independent recognition, it creates less distorted frontal view images by view morphing or homography matrix calculated by the position of the license plate and extracts the keypoints from them. Our method enables to recognize up to 60-degree angle with high accuracy due to the less distorted virtual frontal images.

This paper proposes a novel view independent vehicle make and model recognition method(VMMR). Our system identifies the make and model from the variety of viewpoints while the conventional methods for VMMR work only for the fixed frontal or rear images. In addition, it needs only the 2D images not CAD data for database. To solve the alignment issue, our method uses SIFT, that has scale and rotation invariance. For the view independent recognition, it creates the more realistic and less distorted frontal view images by view morphing and extracts the keypoints from them. Our method enables to recognize up to 40- degree angle with high accuracy due to the less distorted morphed images. Our method can be extended to the other product model recognition.

elighting techniques often require knowledge about the shape and the reflectance property of a target object and the light environment. The quality of the relighting is highly dependent on the normals accuracy of the object since they are used for the computation of the illumination. We propose a new relighting approach for arbitrary shaped objects with dynamic light environment using an RGB-D camera such as the Microsoft’s Kinect. The generated depth map is useful to estimate the normals of the object, but it’s inaccurate because of the noise such as discrete depth values or missing data. We focus on the depth map modification to segment the accurate object region, and normal estimation for relighting. Our implementation achieves a relighting at 10fps with GPU. Therefore, our relighting system can apply for a moving object and a dynamic light environment in real time.