In this study, for the purpose of taking selfies that can be applied even when there are multiple participants, we propose a system for generating a perspective projection image from omnidirectional camera images in which all the participants are within the angle of view and have a specific facial expression as much as possible.

View full abstract

Dynamic visualization, which is a feature of digital museums, can be realized through differential equations. One method of solving this is the trajectory of state vector in a multidimensional vector space. An example of applying this method to Van der Pol equation, which is the basis of nonlinear self-excited vibration, has been introduced and explained.

View full abstract

A negative correlation between the development of lifestyle diseases and protein intake is known in contemporary Japan, where a substantial proportion of the population is concerned with health issues. In this study, we aimed to develop a chicken meat product with enhanced functionalities to provide the society with a solution to increase protein intake. Specifically, we developed a centrifugal accelerator using space technology to breed chickens under a high-gravity environment; thus, we successfully improved the quality of chicken meat by utilizing the effect of the gravity response of the organism to enhance functionalities of its meat, that is, anti-gravity muscles. The chicken meat obtained in this study was leaner than that of chickens bred using conventional methods. Additionally, the meat had a firm and resilient texture that added more value to the meat quality of “jidori” chicken breed used in this study. We have succeeded in non-destructively estimating the hardness of chicken meat reared under this high gravity condition by analyzing image measurement data. This study showed that it is possible to sort high value-added chicken meat without contact.

View full abstract

Our research group is conducting studies on Computer-Generated Holograms (CGH) for volumetric hologram printers. In the process of creating CGH, we convert objects represented by polygons in computer graphics (CG) into point cloud data to compute the interference fringes of the hologram. In conventional research, when converting CG objects into point cloud data, the hidden surface removal process was performed on the CG object from a single viewpoint, and this data was supplemented with the calculation range to create the interference fringes. Consequently, when observing the created interference fringes, there were issues of hidden surfaces missing or overlapping as the viewpoint moved. In this research, we propose a method to improve hidden surface processing by generating point cloud data from multiple viewpoints and switching to point cloud data corresponding to the viewpoint position during hologram computation. We implemented and verified the program for this method.

View full abstract

In this paper, we shall report that, to improve the stability of the mist screen, a spatial screen employed the newer structured nozzle to obtain mist-flow with enhanced directionality. As this result, we confirmed that a performance of more stability and highly bright projected images is carried out by this new system.

View full abstract

Improvement of work efficiency is required in various fields such as entertainment, learning　and physical training. Several kinds of software and equipment have been developed as tools to improve work efficiency. On the other hand, work efficiency and human time perception are thought to be closely related. Although many studies have been done on methods to manipulate human time perception, there is no study on the relationship between human time perception and work efficiency. The authors focused on the impact of accelerated time perception on human work efficiency.　We have developed the environment in which time in virtual space is presented faster than in real space with external visual and auditory stimuli as methods to manipulate human time perception. This paper reports the results of a basic study on the effects of time acceleration in virtual space on participants' workload, workload limits, and fatigue levels.

View full abstract

Exercise is one of the effective means for maintaining health, but strength training plays an especially important role. However, it is difficult for many people to continue muscle training, and continuity has become a major issue. The authors are conducting research aimed at realizing a method to improve the sense of accomplishment during training and increase the sustainability of motivation for muscle training by using a three-dimensional virtual space. By taking ``push-ups'' and ``squats'' as typical muscle training tasks, and using the sense of accomplishment scale and Likert scale as methods for evaluating the sense of accomplishment, we will proceed with experimental verification of the main factors that lead to a sense of accomplishment. We are trying to propose a method to improve the motivation to continue muscle training. This report provides an overview of the experimental system, as well as the results of a basic study on the relationship between changes in the appearance of avatars in virtual space, a sense of accomplishment, and increased motivation to continue muscle training.

View full abstract

We propose a memory aid application aimed at helping prevent forgetfulness and loss at home. Specifically, we have developed a memorandum that takes photos with a smartphone that is used daily and stores them together with notes that contain related information.

View full abstract

The importance of digital archives and digital exhibition and viewing experiences has been reaffirmed. We have developed a virtual museum with automatic sorting for use by both novice art students and those with curatorial-level knowledge. This time, we limited the selection to Van Gogh's paintings. By using ChatGPT, a generative AI, the user interactively communicates his/her requests to the AI in writing, the AI identifies the paintings to be exhibited based on the AI's responses, and exhibits them in a virtual space of computer graphics. In this paper, we compare this application with other virtual museums and show its superiority in terms of freedom of operation, number of languages supported, curation, and conversational interaction. Finally, as future prospects, we mentioned the expansion of evaluation experiments to the general public, the improvement of automatic selection, the increase in the number of supported works, and the improvement of the UI.

View full abstract

The Rakuchu-Rakugai-zu VR contents have added various functions, but since there is no sound, the VR experience is silent and you cannot feel that you have entered the world of the Rakuchu-Rakugai-zu. To solve this problem, walking and environmental sounds were added. In addition, the names of some people and the names of stores are displayed, but the names of buildings other than the names of streets and stores are not displayed. A function to display the name of the street and the name of the building was added so that the user can see which street he/she is walking on and what name of the building he/she has entered during the VR experience.

View full abstract

In the creation of animation models based on character illustrations, such as VTubers, it is necessary to divide the target illustration into parts for each part, such as hair, face, and body. Usually, this process involves using graphics software to divide the parts layer by layer. However, in the creation of complex animations, the parts become more detailed and numerous, making the process time-consuming and labor-intensive. Therefore, this paper proposes a method to improve the accuracy of automatic partitioning using U-net, which was proposed in a previous study, by combining post-processing based on image processing.

View full abstract

Research on people counting systems using image processing is actively conducted; however, the method that can deal with people’s shadows and changes in light conditions has not been established yet. Therefore, in this study, we propose a method of people counting system, which can reduce the influence of shadows appearing on bird’s-eye movies. Local Binary Patterns (LBP) features are computed on the input images, and changed pixels are detected based on the changes of LBP features between frames. Subsequently, region segmentation is performed on the input images using Simple Linear Iterative Clustering (SLIC), and the possibility of a person is determined in each region based on changes in brightness and LBP features. Finally, the person area is obtained by applying a region expansion method using the background difference and RGB changes, which original point of the method is the pixel detected as a person. Moreover, if the person area is composed of multiple people, each individual person is detected using the movement trajectories of persons and the K-means method. At last, the number of detected persons and their passing directions are determined. As the result of the experiment with real videos, the F-score of person counting was 0.96, and passing directions for all individuals were correctly determined.

View full abstract

In fields such as CAD, computer graphics (CG), and animation, research on surface processing involving curve intersections and deformations is actively conducted. When dealing with curve processing, the issue of interference (intersection or distance intervals) between curves and objects/(line segments) is crucial. NURBS (Non-Uniform Rational B-spline) curves are valuable for curve processing due to their expressive power, precision, and usability. In this paper, we propose a method for calculating the intersection and shortest distance (clearance or gap) between NURBS curves and line segments.

View full abstract

The concept of “Modulated Reality (MoR),” which modulates space through the interaction between a real object in real space and an aerial image displayed using an AR device placed in front of it, has been proposed. However, the optical characteristics necessary for implementation have not been clarified, and at present it is not easy to systemize and utilize MoR as a new 3D display method. In this study, we will establish the optical characteristics necessary for systemization while clarifying the relationship between the two basic aspects of the interaction between the real object and the aerial image in terms of luminance. In evaluation experiments, we clarified the relationship between absolute luminance and depth perception. 3D real objects are perceived differently from their actual shape and made to feel deformed in images.

View full abstract

A glassless 3D display is expected to be used not only in industrial applications but also in personal fields. We have previously proposed a concept for a personal glassless 3D display system using retroreflection. In this study, to realize this concept, we understand the optical properties of retroreflection and establish specific methods for its application. Using general optical materials and optical systems, we identified optical properties, particularly luminance, diffusion, and retroreflection characteristics, as a function of the projection angle. We formulated specific prototype specifications.

View full abstract

A state of emergency was declared to prevent the spread of the COVID-19 virus. As a result, it became difficult to hold traditional live shows and events, and online live shows and events have rapidly spread. However, it is difficult to experience the atmosphere of a live show online, and it is difficult to feel the afterglow because fans are immediately returned to reality after the show ends. In this study, to solve these issues, we develop and investigate a system that aims to recreate the atmosphere of a face-to-face live show by setting up a virtual lobby before and after the show, creating a place where fans can interact with each other, and creating a space that recreates the actual live show.

View full abstract

It is realized that real-time 2K (1,920 x 1,080 pixel) full color rainbow hologram display with a transmission liquid crystal panel which is taken from low-cost video projector. The liquid crystal panel is 70 mm (2.76”) diagonal size with RGB color filter and polarizers. Therefore, full color rainbow hologram can be reconstructed with a white light LED and a lens. Since the panel has the color filter, color will not change when viewer moves up and down, unlike normal rainbow hologram. Although the pixel pitch of the panel is 31 µm that does not provide binocular observation, it is suitable for near-eye or eye-tracking type display. With real time hologram generation program running on a note PC, holographic video display can be realized easily.

View full abstract

In the past, the degree of progression of cataracts has been evaluated based on the pigment in the lens. However, this method relies on subjective evaluation and does not take into account the amount of cataract opacity, which may lead to inconsistency in the evaluation. In this study, we propose a method to locally extract the central region of the lens from images of cataract patients. We also present a clustering method based on the color information obtained from the central region and show the validity of the method for image classification results.

View full abstract

In Japan, there are many narrow roads, and it is said that acquiring a sense of vehicle width and improving vehicle operation accuracy are important for safe driving. However, currently, these skills are typically acquired through accumulated real-world driving experience. With recent advancements in simulation technology, it is considered possible to efficiently acquire a sense of vehicle width. Nevertheless, the driving simulators widely implemented in driving schools today are primarily used for basic driving lessons and traffic safety education through virtual accident experiences. Therefore, we hypothesize that driving simulators could also be effective in acquiring a sense of vehicle width and improving vehicle operation accuracy, and we have begun basic investigations towards developing a driving simulator suitable for this purpose. According to insights from the field of cognitive science, it is important to reduce the prediction error between the brain's prediction of the vehicle width and the actual vehicle width, and to minimize the prediction error between the predicted and actual passage trajectories through steering operations. To accelerate the minimization of these prediction errors, we have devised a new display method that visualizes the actual vehicle width by rendering the vehicle body transparent, showing only the tires and side mirrors, and displaying the ideal tire trajectory as teacher data on the road. This report presents an overview of the devised VR driving evaluation system and the results of effectiveness verification using this system.

View full abstract

Games and video experiences in virtual space using head-mounted displays were initially provided on the assumption that the postures of users in real and virtual space would match, however, the postures of users while using these services have become more diverse, with users lying down in real space and running in standing position in virtual space. It is still unclear how the sense of immersion changes due to the mismatch of postures between real and virtual spaces. The authors have constructed an experimental environment in which participants can change their posture in the real space while maintaining an upright posture in the virtual space and have been experimentally verifying the effect of the difference in posture between the real and virtual spaces on the sense of immersion. In this paper, we report the results of our evaluation of how the sense of immersion changes when the difference in posture between real and virtual space is changed step by step.

View full abstract

By constructing an image bump evaluation system in which image bumps used on actual roads are formed on a virtual road and driving experiments are conducted, it is possible to (1) narrow down the various designs and drawing positions to some extent and (2) grasp the effects of environmental changes such as environment (residential areas, depopulated areas), weather conditions (sunny, rainy, snowy) and time changes (daytime, dusk, nighttime). We have developed a “basic system for preliminary verification using virtual space,” which will lead to a huge reduction in time and labor, as well as to the realization of accurate analysis under uniform conditions. We report here on the prototype and the evaluation results.

View full abstract

This study presents the development of a prototype finger recognition system aimed at improving typing skills. The system utilizes MediaPipe Hands to detect hand landmarks in real-time, implementing keyboard calibration, real-time finger recognition, key press detection, and visualization of typing status. Key features include individual tracking of all ten fingers with left and right hand differentiation, mapping of pressed keys to specific fingers, and real-time visualization of typing status. This paper describes the system's architecture, implementation methods, and functional overview.

View full abstract

A promising approach for low-cost implementation of traffic volume monitoring is to utilize already deployed Closed-Circuit Television (CCTV) cameras with image recognition technologies. However, the problem is that the counting accuracy is degraded when vehicles are occluded by other vehicles or road lighting fixtures. In this study, we propose a traffic measurement method that enhances robustness against occlusion by focusing on the property of following vehicles to the vehicle in front and predicting the position of occluded vehicles according to the car-following model of traffic flow theory. Experiments using CCTV camera images showed that the error rate decreased by 1 to 11 points in situations where the road was congested and heavily occluded.

View full abstract

Dairy farmers are busy managing the health of their dairy cows, and there is a need to automate individual dairy cow management. Dairy cow identification technology is important for automation, but conventional methods using deep learning have not been sufficiently accurate because it is difficult to collect training images of dairy cows in complex postures from various directions. In this paper, data augmentation by automatic generation of individual dairy cow-specific spotted patterns is investigated with the aim of improving the accuracy of individual dairy cow management using surveillance cameras. Specifically, data expansion is carried out to generate images with interchangeable pattern of spots using DG-Net, an image generation method for person re-identification, and the effectiveness of the data expansion for individual dairy cattle identification is demonstrated through simulation experiments using 3DCG. In addition, the relationship between posture change, the number of generated images and accuracy is also evaluated.

View full abstract

In this paper, we propose a novel image generation technique that enables the creation of designs closely matching the user's imagination in the partial editing of objects in images. Image generation using partial editing has seen various research and application developments, particularly in applications such as interior design coordination and Virtual Try-On systems in XR (Extended Reality). Despite the significant improvements in accuracy and quality brought by the advent of diffusion models, it remains challenging to faithfully materialize user-imagined designs on images while preserving the texture and structure of objects, thus providing a satisfying level of generation quality. To address this issue, we introduce a mechanism within the image generation process, which includes Stable Diffusion, aimed at overcoming the aforementioned challenges. Specifically, we create two types of feature representations: one focusing on the texture and object structure and the other on design patterns. By introducing these features into the intermediate layers of the image generation model, it becomes possible to ensure the integrity of the object during the reverse diffusion process. This paper demonstrates the effectiveness of the proposed method by applying it to the task of furniture texture modification. We expect that the proposed technique will contribute to the advancement of image synthesis methods using generative models.

View full abstract

Although object detection technology with cameras holds promise for various applications, it incurs dataset creation costs to train new models where general-purpose models are ineffective, such as in industrial settings. We have previously developed a semi-automated annotation framework that employs optical flow and representation learning techniques to reduce human effort significantly. However, it was prone to unintended annotation omissions compared to manual annotation. Therefore, this study introduces image synthesis to overcome these annotation losses. The proposed method synthesizes natural-looking images without annotation losses by placing labeled foreground segments at their original positions on targetless background frames collected with the same fixed-point cameras. Evaluation with video footage from wide-angle cameras on a logistics warehouse ceiling indicated that improved dataset quality led to higher model performance.

View full abstract

Computer vision (CV) and computer graphics (CG) are beginning to be used in the forensic science domain. In a previous study, a method was proposed for estimating a real-scale 3D human body shape model from a single person image based on 3D camera calibration. Although estimation has become possible even in the non-straight posture conditions such as walking and running, the accuracy of analysis under non-straight posture conditions other than walking and running has not been clarified. In this paper, for aiming to clarify this issue, we newly analyzed and evaluated the proposed method under four non-straight posture conditions: sitting, crouching, and holding objects. As a result, we were able to evaluate two points: whether it is possible to analyze a 3D human body shape model for the four posture conditions, and the average analysis accuracy of height and weight estimation. Based on the results of this paper, careful practical usage of the proposed method is required.

View full abstract

This paper proposes the versatility of a multi-camera tracking method for dairy cows using rotated rectangles and the effect of changes in angle of view on tracking accuracy. Conventional rectangles are aligned with the horizontal and vertical axes of the cameras, and the detection rectangle contains many extra background regions, which contributes to the reduction of tracking accuracy. Therefore, we have improved the tracking accuracy by using the previously proposed rotated rectangle to more accurately determine the position of dairy cows. In this paper, we investigate the versatility of the system when using different barn data sets. We also aim to reduce over-integration of tracked dairy cows by changing the angle of view of each camera, and to examine the effect of setting exclusion zones on tracking accuracy in multi-camera tracking.

View full abstract

An event held for the purpose of social interaction is recorded with a camera. The obtained footage is then analyzed to estimate mouth movements and gaze directions. A temporal analysis is conducted to estimate the participants' levels of interest.

View full abstract

We will collect comments on waterside spots from websites that contain reviews of tourist spots. In order to verify the difference between the sensibility that people have and the sensibility that AI perceives, we will compare the comments with the verbalized images of waterside spots using ChatGPT4o. We extracted sensory words from both images using a linguistic analysis tool and compared them. We also focus on the co-occurrence relationship of the sensory words.

View full abstract

We are developing a seismic simulation MR Glass that can routinely monitor the seismic readiness of furniture, storage cabinets, household items, wiring and other objects in a realistic indoor environment in terms of their condition and organisation, and periodically check them against seismic simulation. This report presents the results of a study of basic object condition detection and seismic simulation.

View full abstract

In recent years, how to coexist commercial billboards with the townscape has become an important theme for the new design of billboards in townscapes. Therefore, in this study, we cut out certain advertising billboards into line drawings and change the thickness of the line of the sign's pattern and letters to pursue a sign design that fits the landscape.

View full abstract

The application of the arithmetic coding to image signals has started from binary images and extended to multi-level images and video, nowadays. For the application of arithmetic coding to multi-level sources, usual conventional method has been the usage of binary arithmetic coding by decomposing multi-level sources into plural binary sources. We have proposed multiplication-free multi-alphabet arithmetic coding (MFMAC) in which calculation of the width of each symbol will be done in multiplication-free way. We have studied the basic consideration of MFMAC, by assuming the source model, in which, the probability of the ranking symbol will be decreased by a fixed rate R, and the learning method to guess the value of R by estimating the probability of the most frequent symbol, and the acquisition way of occurrence ranking information. In this report, the outline of MFMAC will be described based on the study result, and the performance for its most weak sources will be checked.

View full abstract

Sign language recognition technology that uses skeletal information estimation from video images has been reported. We have applied this technology to develop a learning support system that scores sign language movements based on the likelihood of sign recognition and provides feedback to sign language learners. However, conventional sign language learner support system was incorrect when an incorrect sign language movement was made. In this report, we create multiple body part information missing identification models trained from data lacking skeletal movement information for the right arm, left arm, etc., and propose a technology to estimate the body part of incorrect sign language movements. We implemented the proposed system and conducted evaluation experiments using movement data from sign language beginners to confirm its effectiveness.

View full abstract

When light falls on a surface, various patterns appear due to reflection. We are interested in studying the diversity of these pattern appearances from the perspective of mathematical surface theory and singularity theory. In this study, we consider the intensity distribution on a smooth surface and investigate the relationship between its local singularity types and the curvature of the surface. Additionally, we are studying how the singularity types of the intensity distribution change when the light source is perturbed, from the perspective of the ″versality″ of singularities. In this presentation, we will introduce the above mathematical results along with specific examples of surfaces.

View full abstract

Modern companies place importance on ″communication skills″ when hiring new graduates, but face-to-face communication has been decreasing due to the development of information technology. In this study, we focus on group communication. We will develop a training system to improve each of the three evaluation items, ″facilitation skills″ , ″communication skills″ , and ″assertion skills″ , by using gaze estimation, voice recognition, and questionnaires. In addition, by using VR in the development, we aim to improve the skills effectively through practical learning.

View full abstract

As members of society, they have far more opportunities to communicate with their superiors, give presentations, and speak in meetings than when they were students, and therefore, presentation skills are in great demand. Since a large percentage of people are not good at public speaking, this study aims to improve presentation skills by focusing on voice, facial expression, and eye contact, evaluating presentations, providing feedback, and then using VR training.

View full abstract

Logistics services such as catering have become a social issue due to a labor shortage. Under these circumstances, several demonstration experiments have been conducted to realize catering using drones. The conditions for catering require basic functions such as the ability to reliably deliver to a designated point and the ability to transport contents without spilling them. In this research, we developed communication functions to create a sense of familiarity with catering drones in a limited communication environment. And basic experiments were conducted.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

Many communal facilities such as restrooms and changing rooms, which are frequently used by many people, face numerous challenges for both users and managers．Users cannot determine the availability of the facilities without going to the location．Meanwhile, managers struggle with a lack of usage data, hindering smooth operations such as maintenance and cleaning．To address these issues, we developed a utilization visualization system using Wi-Fi packet sensors．This system collects radio waves from Wi-Fi devices, and despite the randomization of MAC addresses, it can estimate and visualize the number of users．To evaluate this system, experiments were conducted at two communal restrooms with different layouts at the Hirakata Campus of Osaka Institute of Technology．The experiments compared the user count estimated by our system with those manually recorded．The results showed a high correlation of about 90% at both restrooms, with an average error of 0.52 and 0.76 users, respectively, confirming the effectiveness of our system．

View full abstract

In learning sign language, it is quite difficult to evaluate the validity of the memorized sign language motions when there are no sign language users nearby. As a means of solving this problem, the authors are developing a sign language review tool in which users can reproduce their sign language motions in front of a camera, and the results are judged by AI and fed back to the user. The review tool compares the video of the sign language motion acquired with an emphasis on efficiency to create a classification model of the motion with the video data acquired with an emphasis on automatic judgment of the start and end of the user's sign language motion in the sign language review tool. In this case, there are differences in the range of vision and the time segment to be classified between the two, which affect the classification performance. In this report, we describe a method for absorbing differences between the video data used to create the classification model and the video data of the user's sign language movements when using the sign language review tool, and the results of confirming the effectiveness of the proposed method.

View full abstract

It is clear that insects rely heavily on vision when preying on moving prey, escaping from natural enemies and finding nectar in flowers. The compound eyes of insects have an optical structure that differs significantly from that of the human eye. The structure of the compound eye has long been understood. However, not many studies have shown what images of the external world are captured by such structures. This study proposes a method for estimating the image seen by the compound eyes of diurnal insects. We focused on the relationship between the viewing angle of each individual eye and the angle between neighboring individual eye lines of sight. It predicts how the view differs from a distant view to a very short close-up distance. Simulation experiments were conducted on a CG model of the outside world, which imitated a flower field, to estimate the outside world image.

View full abstract

QR Code uses RS code to provide an error correction function and achieves correction performance up to its correction limit． Its maximum correction rate is about 30% of all data code words． However, the correction rate of about 30% is limited to correcting data code words and does not address errors in finder patterns and format information that are essential for reading data． It cannot handle a wide range of burst errors． Therefore, finder patterns and format information are added and arranged in a dispersed manner while maintaining compatibility with reading by existing reading devices． Furthermore, the data code word portion is duplicated． An shell structure is used for the additional arrangement of finder patterns and format information． A chimera structure is used for duplication of data code words． This makes it possible to handle burst errors of up to 60% for 2d symbol area．

View full abstract

In recent years, Virtual Reality (VR) market has been expanding due to the influence of the metaverse and the proliferation of home VR devices. However, there are few studies on the effects of VR environment on the body, particularly on peripheral fatigue, which has some unclear effect points. One method to investigate peripheral fatigue is to analyze muscle fatigue through changes in electromyography (EMG). We demonstrated the potential for muscle fatigue relief when viewing forest VR environments while fast walking on a treadmill, as indicated by EMG. In this study, we investigated the increase in muscle fatigue, which is the opposite of fatigue relief, by inducing fear. Where four types of VR environments were created: a normal city environment, and three fear-inducing environments - height, the sea, and dark forests. Five subjects walked in these VR environments using a head-mounted display. During the VR experience, frequency waveform data was obtained from EMG sensors attached to the rectus femoris muscle. Three types of frequency analysis were employed to investigate muscle fatigue: fast Fourier transform (FFT), integral values, and continuous wavelet transform. The results suggested that VR imagery inducing fear tends to increase muscle fatigue in the rectus femoris muscle.

View full abstract

When a large-scale disaster such as the 2024 Noto Peninsula Earthquake occurs, municipalities and other relevant authorities conduct damage assessment surveys on residential houses. These surveys are primarily conducted through on-site inspections and are expected to be prolonged. The damage assessment surveys are carried out by evaluating the exterior, tilt, and specific parts of the buildings. Previous research proposed a method to estimate the tilt of buildings by generating 3D models using SfM/MVS from images taken with a fixed smartphone and by utilizing the accelerometer values stored in the captured images. In this study, we aimed to extract the tilt of objects during handheld shooting based on the methods employed in previous research. Experiments were conducted to verify the effectiveness of both fixed and handheld shooting using the previously proposed method. It was found that there was room for improvement in the effectiveness during handheld shooting. The accuracy of the tilt estimation during handheld shooting was improved by combining images from the captured image sets, generating multiple models, calculating multiple estimated tilt values, and enhancing the centrality through statistical processing. Additionally, the potential for further accuracy improvement was evaluated and examined by applying a voxel filter using mesh resolution to the 3D models.

View full abstract

Breast Density, a medical term, indicates the ratio of glandular tissue to fat region and is classified into four types. Currently, the determination of breast density is performed by qualified image interpreters. However, since the judgment of the boundaries of the mammary gland area depends on subjective factors, the assessments can vary between interpreters. In this study, we are working on developing the automatic classification of breast density using VGG16, one of the image recognition models. We also employ Grad-CAM to investigate how specific regions contribute to classification and their relationship to the mammary gland area, aiming to understand the causes of misclassification.

View full abstract

The visually impaired person obtains the desired e-book with the help of an audio guide and reads it out loud by a screen reader. This section clarifies how the visually impaired, who owe their use of e-books to their personal abilities, use e-books.

View full abstract

In recent years, the development of VR (Virtual Reality) technology has made it possible to experience an unprecedented level of visual presence. Anyone can easily become a performer, and by entering the metaverse space, spectators can watch someone perform. However, despite the rich visual presence, the widely used sound technology in the metaverse space lacks the immersive feeling of being in the moment, as it does not provide a three-dimensional sound experience. This paper proposes a system that converts sound into an immersive surround sound environment during transmission within the network to construct a three-dimensional sound space in the metaverse. Specifically, we developed a prototype system that integrates a server for multi-channel surround conversion, an RTSP distribution server for multi-location distribution, and cooperation with VRChat, one of the metaverse platforms, enabling anyone to experience this sound environment in the metaverse space. We evaluated this system through public events and experimental exhibitions.

View full abstract