ABSTRACT. The highest resolution of digital elevation model (DEM) data published by the Geospatial Information Authority of Japan is 5 meter and it is provided as DEM5A. Since the DEM5A data is generated based on data obtained by aviation laser surveying, it can not obtain correct elevation on the water surface. Therefore elevation value for every DEM's cell which is included in the water surface is assigned value of -9999. We can not correctly show the elevation of the water surface by directly using the DEM5A data. In order to solve this problem, a method for classifying water areas into three types according to the water surface characteristics and assigning an appropriate elevation to each type was proposed. However, this method requires some user interactions to a water portion detected as a river. Specifically, this method requires the user to subdivide the river portion into several fragments. As a solution of this problem, we propose a method to automatically assign an appropriate elevation to every DEM cell included in a river portion.

ABSTRACT. In recent years, services for automatically generating facial images using AI have increased, but it was not possible to generate differential images such as facial expressions and poses necessary for game production, and manual correction was required. Therefore, in this research, we propose a method to automatically generate facial images with pose and facial expression differences by mining the latent space based on the pose analysis and facial expression analysis of the generated images using StyleGAN2. We develop an evaluation system based on the proposed method and describe the evaluation of its effectiveness.

ABSTRACT. For safe remote control of large robots, it is important to see the environmental information intuitively from scenic images around the robot. We generate environmental background images by concatenating multiple images with in-vehicle fisheye cameras. For image concatenates, camera extrinsics are estimated by several markers on the objects at infinite distance. As the poses of cameras on this system vary with long-term use, camera extrinsics have to be estimated at each operation. It is, however, hard to prepare the marker objects at an infinite distance during operation. Here, we proposed the estimation technique to acquire the camera extrinsics using surrounding markers on environmental structures. We show the effectiveness with experimental results on 2-fisheye-camera systems using surrounding street images.

ABSTRACT. We have developed a high-quality, real-time 3D videoconference system with the simplest configuration of stereo cameras and an autostereoscopic 3D display. The system’s effectiveness as a 3D videoconference system was discussed by conducting 2D videoconferences with the system for multiple users. However, quantitative evaluation of our videoconference system could be more challenging. Therefore, we utilized video-based gaze tracking to analyze the frequency of eye contact during meetings, demonstrating the effectiveness of this system in video conferencing. We validate the effectiveness of this system in video conferencing through an analysis of eye contact based on the results of gaze tracking and subjective evaluations obtained from surveys.

ABSTRACT. What is the significance of digital information that is invisible to the human eye? There are several benefits of the invisible digital information. For example, invisible QR code can be considered to improve the security of products without compromising their design. We have succeeded in making achromatic invisible QR code on a niobium substrate by thin-film interference of niobium oxides using side-by-side additive color mixture. Although this digital information is invisible in the visible light range, but detectable in the infrared light range. This technology has a potential to be applied to anti-counterfeiting and traceability.

ABSTRACT. Japanese manga is highly acclaimed worldwide and entertainment for all generations. One of the characteristics of manga is the title logo, which uses various fonts and designs that match the contents of the manga. This study uses a rough set to analyze the characteristics of title logos for each manga genre, such as shoujo manga (for girls) and shonen manga (for boys). This will allow us to know what kind of title logo designs are preferred in each genre and is expected to serve as one indicator of title logo design.

ABSTRACT. This study introduces a visual simulation method for the sublimation of dry ice in water, utilizing flame shapes obtained from combustion simulations. While 3D graphics often depict fog resembling that produced by dry ice, instances focusing specifically on the sublimation phenomenon itself have been overlooked until now. Our research replicates the Leidenfrost effect in underwater dry ice using flame shapes and applying smoothed processing to depict distinctive changes in bubble shapes near the submerged dry ice. This method enables the 3D graphics reproduction of events, from the introduction of dry ice into water to the overflow of fog due to sublimation.

ABSTRACT. In recent years, research in the field of three-dimensional computer graphics (referred to as 3DCG or CG) simulation has advanced, with increased attention not only limited to physical or natural phenomena but also towards CG representations related to food. This paper discusses the CG simulation of interactions between spaghetti and sauce. However, interactions between thread-like objects such as noodles and viscous fluids like sauce pose a challenging subject due to their mutually complex behaviors. While there are several physics-based approaches to these, their intricate algorithms and heavy computational costs make them unsuitable for all scenes. This paper aims for application in the entertainment field and achieves a simpler computational approach, lighter than physics-based methods, by simulating interactions of both thread-like models and liquids using a position-based method.

ABSTRACT. We aim to verify how user experience changes when hit stop duration, one of the elements of GameFeel, is altered based on gaze information. A preliminary experiment identified the comfortable range of hit stop durations, and main investigation explored how varying this duration within the comfortable range affected the response based on gaze duration. As a result, designing hit stop duration that corresponds to staring duration is important to improve GameFeel.

ABSTRACT. Coloration with a sense of color unity is important for attractive illustration expression, but this is difficult for beginners. However, this could be achieved by proceeding with expert’s process of color planning, color scheme, and color adjustment in a predetermined order. Therefore, this study proposes a digital illustration system for beginners, that the system presents color scheme examples according to the user’s color plan to give them a concrete image of the completed image, and then they color the image while reviewing and adjusting the coloring according to the instruction messages, thereby enabling coloring with a uniform sense of coloring.

ABSTRACT. Japanese animation features a variety of unique expressions such as exaggeration and distortion, and animators spend a long time acquiring the skill to depict such expressions. Observing good examples is helpful for acquiring such skill,but finding appropriate examples for a given situation is difficult. In this paper, we discuss the way to provide a method that display past drawings as references for animators to draw. Animators input what is related in a given shot, and our method retrieves past drawings as appropriate examples of how to draw.

ABSTRACT. Human speech organs are modeled using computer graphics modeling software. Most of speech organs are inside body therefore shapes and motions of speech organs during speech producation are invisible and hard to be understood. In this study, speech organs are modeled both as computer graphics and as phyisical figure models. At shis time, upper and lower jaws including teeth, tongue, lips, vocal chords, etc. are modeled using computer graphics software. As reference models of speech organs, MRI (magnetic resonance imaging) data and anatomical drawing data are used. The models are developed as contents of educational materials to teech human speech production. Also, the modeling data are transfered to make pysical figure models for 3D printers. They are also used for a museum exhibition.

ABSTRACT. This paper presents a new type of extended reality (xR), namely modulated reality (MoR), which fuses real and virtual objects through depth fusion. With this fusion, even though the real object can be seen directly with the naked eye, it is perceived as a differently shaped object from reality. Through experiments using a printed image as a real object and a head-mounted display to display virtual objects, we demonstrate the feasibility of MoR. The experimental results showed that the depth of the fused image depends on the luminance of the virtual object displayed in front of the real object, similar to conventional depth fusion; this implies the possibility of the shape modulation of a real object.

ABSTRACT. An event camera has a vision sensor that asynchronously detects pixel-by-pixel brightness changes. Compared to the conventional RGB camera, it provides better detection accuracy when objects move fast or are in dark areas. There are several studies in which a 3D pose is estimated from event-based video. However, joints with slight motion are not displayed, leading to a loss of pose estimation accuracy. In this paper, we propose a method to estimate joints that cannot be detected with a single frame. We use past and future frames as well as current frames for estimation. From the experiment, we confirm that the pose estimation accuracy is stabilized and improved.

ABSTRACT. Livestock counting or recognition is a crucial process of cattle management in every farm worldwide. Farmland owners typically use drones to capture instances or footage of livestock, especially when the livestock count is massive. Livestock recognition of drone footage requires many variations of aerial images or videos of livestock for training, which are generally insufficient in real life. Our paper demonstrates a method of data augmentation using 3D models and scenes rendered for the training dataset of livestock recognition. Experimental results in our research convey a satisfying detection result, validating the use of 3D-rendered models for data augmentation.

ABSTRACT. During disasters, it is crucial to assess evacuation centers' situation quickly. However, obtaining information through human patrolling is time-consuming and inefficient. Furthermore, locating evacuees who are not in evacuation centers can be challenging. We developed the Information Sharing System for Evacuation Center Management (ISEM) to address this issue. ISEM can share information on evacuation centers in real-time using smartphones and other smart devices.