With the increasing computing power of computers and the increase in data scale, artificial intelligence has entered a new era of prosperity and development. However, since there are very few artificial intelligence applications in the animation industry, there is still a lot of room for automatic generation of animation covers and intelligent editing of animation highlights. In this paper, we proposed an intelligent analysis algorithm for animated characters based on target detection. This is a multi directional improvement of the SSD target detection algorithm. In the initial stage of the experiment, increase the number of data and select a pre network structure with high feature expression ability. Then, the performance is improved by setting an appropriate default box shape according to the annotation characteristics of the anime character. The accuracy of the results was improved by 10% compared to t he results at the beginning of the experiment.

View full abstract

In this work , we propose a method to calculate the overlap rate between frames using motion vectors and optical flow. In this method, the amount of frame movement is calculated from the amount of change in each vector detected. The accuracy of each vector was evaluated by experiments using moving images with different set values, and the overlap rate of 3D model reconstruction was examined.

View full abstract

In this paper, we propose a method to express the shade that be created on a virtual object placed in the environment by estimating the light source in the real environment using RGB-D sensor images. This method estimates albedo by performing clustering using the K-means method for the hue of color image and the normal map calculated from depth image. The virtual object is shaded by the irradiance map of the environment from this value, the diffuse components value, and the normal map. In this way, we aim to keep the optical consistency of virtual objects in augmented reality and improve the sense of reality without using objects or devices such as mirror spheres for estimating the light source near the observation target.

View full abstract

In this study, we developed a system that replaces real objects with virtual objects of the same category and displays them. By combining general object recognition and inpainting into AR, real objects are visually erased and only virtual objects of the same category as real objects are displayed. As a result, the real object appears to be replaced by a virtual object of the same category on the screen. The system obtains the category and area of the object to be removed by general object recognition, sets an image that the object is erased by Inpainting as the AR background, and places virtual objects of the same category as the recognized category. In some scenes, real objects block the plane where the virtual object is placed, so the depth image is modified to recreate the hidden plane, and the virtual object is placed using the repaired depth image. The system works with a single smartphone equipped with an RGB camera.

View full abstract

In this study, we investigated a method to reduce the discomfort of structural inpainting. Although the structural type is a classical method, it has the advantage of versatility in that it can be applied to images where defects are widely and densely distributed. However, the completion accuracy of the structural method is relatively low, and it causes a great deal of visual discomfort. In this study, we propose a method to reduce this discomfort. The outline of the proposed method is as follows: smooth the image completed by the structural type, search for similar pixels in the source region, and complete those pixels. In our experiments, we conducted visual evaluation and quantitative evaluation by RMSE. As a result, we were able to greatly reduce the discomfort in the completion region while suppressing the deterioration of RMSE.

View full abstract

Inpainting is a technique for complementing a loss region part based on surrounding information. The conventional method is not good at images containing a wide range of thin loss region and thick loss region. In this study, We tried to solve the problem by combining Exemplar_based and Structural method.

View full abstract

In recent years, with the development of paperless documents and optical character recognition (OCR), it is important to binarize document images converted into digital data. However, when binarizing a document, if the document is deteriorated due to stains such as stains or strike-through, or if the document is blurred or missing, over-extraction or area loss will occur. Good binary results are not obtained. In this study, we proposed and quantitatively considered a binarization method to improve over-extraction and domain defects, with the main purpose of verifying the morphology operations used in these degraded document images. In the binarization experiment, the effectiveness was confirmed from the matching rate and recall rate, the F value obtained from the matching rate and recall rate, and the evaluation by RMSE for the image data set provided by DIBCO. In addition, it was confirmed that the threshold value of Kenny Edge used in the above-mentioned method is an important factor.

View full abstract

In recent years, autonomous mobile robots, such as self-driving robots, have attracted much attention, and their practical application is expected. In this report, we proposed a fundamental method for detecting stained areas on the sensor using 3D point cloud processing, and present on the results of a verification experiment in which the method was applied to a 3D laser scanning system.

View full abstract

Registration, which is the process of correctly superimposing and merging multiple 3D point clouds, is one of the fundamental techniques in 3D point cloud processing. Although the method of obtaining the coordinate system transformation parameters from the correspondence between point clouds obtained by feature point matching is often used, it is difficult to achieve high precision registration because of the low repeatability of feature points and the strong influence of matching errors. On the other hand, VKOP (Virtual Keypoint Of Polyhedron), which finds feature points at virtual locations, has been found to have high repeatability in environments containing many planes, but no registration method using VKOP has been proposed. In this paper, we propose a new method to obtain the appropriate coordinate system transformation parameters by brute force trials of local reference frames, considering the nature of VKOP. By comparing the proposed method with the conventional method, it is confirmed that the proposed method can be applied with high accuracy and speed.

View full abstract

In this study, we investigate the relationship between the variability (consistency) of line drawing based on point clouds and the reaction delay RMS (root mean square) / the variability of reaction time during a simulated driving task. In order to examine this relationship, 23 participants performed a line drawing task three times and a simulated driving task, and the correlation was examined. The results suggests that inconsistency of line drawing is positively correlated with the reaction delay RMS / the variability of reaction time where the correlation coefficient was about 0.7. The results of multiple regression analysis show that the consistency of the line drawing can predict the reaction delay RMS / the variability of reaction time (coefficient of determination: 0.50, 0.48). From these results, we conclude that people who tend to change the criteria for drawing lines in a certain point clouds are also likely to show variability in a simulated driving task.

View full abstract

This paper presents a simple approach to encode video sequences by selecting an optimal Intra encoded frame (I-frame). I-frames in Versatile Video Coding (VVC) random access mode influence the bitrate required for the remaining frames in the group of pictures (GOP). Hence, we carry out a full search for the optimal frame in the first GOP and select an I-frame with significant spatiotemporal information. We apply this technique to encode JVET sequences using VVC random access mode. Results show that an average bit rate savings (based on Bjøntegaard Delta metric) of approximately 1.2% on the Y-channel.

View full abstract

The keypoint patch extraction (KPP) is the method that speeds up the 3D point cloud registration by reducing the number of point clouds. However, there is a problem that the success and accuracy of the registration depends on the detection location of keypoints. To solve this problem, a registration method has been proposed which selects the KPPs to be referred to avoid the influence of non-overlapping KPPs in the process. In this paper, we summarized the investigation of the overlap ratio to which the method can be applied and discussed the method by focusing on the KPPs selected.

View full abstract

3D point cloud processing is one of the fields of computer vision, and has been applied to various fields in recent years. There is a problem that the accuracy of 3D reconstruction from video sequences is degraded when there are moving objects in the frame. In this study, we investigate a method to improve the accuracy of the recovered 3D model by extracting and masking the moving objects.

View full abstract

Voting-based methods are not a conventional approach for the registration of 3D point clouds. Nonetheless, recent proposals have achieved outstanding results in low overlapping scenarios around 40% and below. The quaternion is the rotation representation of choice to bin into the voting accumulator that determines the best transformation matrix within their pipeline. Compared to the unit quaternion, the Euler angles are a rotation representation formed by fewer arguments and are easier to interpret. However, the order in which these are applied plays a critical role in the final transformation. In this paper, we evaluate the significance of the order in which the Euler Angles are searched and applied within a voting-based registration framework.

View full abstract

A 3D point cloud is a representation of a 3D space with many discrete coordinate information by laser or photogrammetry. By processing 3D point clouds, we can directly analyze the features in 3D space, and thus various applications are expected. In this work, 3D point cloud data obtained from actual remains are used for shape detection using three kinds of indices (DoN, curvature, and ND-PCA).

View full abstract

This paper proposes the method for shooting and synthesizing ultra high resolution images using a pixel shifted continuous shooting camera and frame selective multi frame super resolution processing. High resolution composite images are reconstructed by multi frame super resolution by selecting the frame image that is not affected by shooting blur or atmospheric fluctuations from pixel shifted continuous shooting frame images. In the experiment assuming the application of a surveillance camera, it was confirmed that a clear ultra high resolution zoom image can be taken with the shooting blur and atmospheric fluctuations removed.

View full abstract

Recently, with the rapid spread of smartphones, opportunities to deal with information related to user privacy are increasing, and it is indispensable to implement a use authentication function into the smartphone. A PIN, a password, a pattern lock, etc. are used as traditional user authentication methods in conventional smartphones. However, these authentication functions have a problem that there is a risk of spoofing by guessing the PIN, the pattern, etc. by another person. Moreover, there is another problem that user authentication is not performed continuously after once the authentication succeeded, so there is a risk that the authenticated terminal will be used illegally. Based on the above background, in this research, we propose a new continuous authentication method for smartphones that considers the user’s usage environment and evaluate its effectiveness.

View full abstract

In recent years, the number of services for viewing video content via networks has increased. Moreover, the number of subscribers to the service has increased rapidly due to the corona virus, and due to the spread of 5G services. Thus, the technologies to reduce the amount of data transmission in video distribution services are required. In particular, the improvement methods of video communication using cache have been studied. In the conventional method, it has been reported that the video management method affects the cache efficiency in practical models such as the number of accesses and fluctuations in quality usage. However, this evaluation is limited to the evaluation of the cache effect within one cache server, and there is a problem in the actual user's video usage model without fully considering the case where the user uses the video while acting. Therefore, this study focuses on the performance degradation of a practical cache-type video distribution system, when a user moves and uses video, i.e., there are multiple cache servers. We propose a new video management method that uses a user behavior model linked between cache servers, and consider its effectiveness.

View full abstract

The opportunities for viewing online video are increasing. However, visually induced motion sickness is a concern, when the viewed video contains screen shake information. The countermeasure is one of the important problem. So far, we have studied a screen shake extraction method using a motion histogram to prevent the sickness. However, there is a problem of the limit of extraction accuracy. Therefore, in this paper, we focus on improving the extraction accuracy. In the motion vector space, we propose a new screen shake extraction method using a principal component analysis and a variability analysis of mean values of motion vectors, and consider its effectiveness.

View full abstract

In this study, we analyzed the data of penalty kick kickers in order to develop a system to support their practice of shooting, which is difficult to be predicted by goalkeepers. The results showed that the angle of the kicker's axle foot and the angle of leg opening were correlated with the shooting angle of some kickers, which was useful for understanding the characteristics of the kicker's own shooting.

View full abstract

Until now, the associative studies have been conducted using paper-based associative studies forms. In December 2019, the Ministry of Education, Culture, Sports, Science and Technology launched the “GIGA School Concept”, and the introduction of ICT in educational settings is being promoted. Therefore, we considered making the associative studies compatible with keyboard input. In the case of the paper survey, the filling time was basically 50 seconds. However, the speed of typing could be different from the speed of writing on paper. Therefore, we examined how much input time should be secured when conducting the associative studies using a keyboard.

View full abstract

The technical word “Soft Computing” is explained simply at first, the foundation of engineering fields, and the applications to architecture , robots, aerospace, transportation and so on are explained the details. The author is expected to be applied to other technical fields, for example drone and communication.

View full abstract

When creating robot parts with a 3D printer, a part with insufficient strength was found, and the strength change was analyzed by changing the parameters of the slicer software. In this study, the analysis was performed by changing the filling rate, filling structure, and number of walls.

View full abstract

Manufacturing the pedal type controller which can perform the movement of the robot only in a human leg part and examining the shape and usability. In the remote control of the humanoid robot, a motion capture device is used as the means that a human being can operate for an original sense. However, I feel that there is a problem for control of the movement to the present thing Because it use some functions of the human upper body in movably controlling robot, and the function of the upper body cannot utilize for other work enough I perform a study to solve it. I examine a shape and the usability of the pedal type controller. In addition, I am going to perform the examination of the usability when I combined a motion capture device with it.

View full abstract

We considered a detection system that uses an infrared camera to install in drone or monitoring camera. YOLOv3 is the object detection algorithm bein g used in this study. A b ear we set as a target is supposed to be a harmful animal that attacks victims in a mountainous area. We considered the effectiveness of virtual datasets using 3D models because it is dangerous to capture harmful animals face to face. The recall score on the F1 score of detecting RGB images using virtual datasets shows 61.3[%] 61.3[%]. In addition to this, the recall score of detecting infrared images by datasets using infrared images shows 98.1[%].

View full abstract

Recently, remarkable advancements in an autonomous mobile robot technology enable increase a situation which humans and robots work in a common space. However, it makes collision risk higher as we enjoy benefit from robots even closer. To avoid this problem, safety robot having capability to forecast human motion and move without clash will be important. We propose a method to forecast a human walking motion using 3D poses estimation and deep learning. In addition, we present what optimizer is better for this task and which is better model training, use 2D datasets or 3D datasets. We collected 3D patterns of motion data from depth camera, Intel Realsense. Also, we considered actually mounting our system on an autonomous mobile robot by using NVIDIA Jetson Xavier NX to obtain data in real time.

View full abstract

Introduction of a project by high school students to make cosmetics using oil from sunflower seeds grown in fallow farmland.

View full abstract

A discussion of the functions of the ZOOM video conferencing system to promote emergent dialogue in learning. An attempt was made to introduce a system that can make effective use of non-verbal communication and co-speech.

View full abstract

Online classes with video distribution, which has spread rapidly with the prevalence of COVID-19 infection, are now routinely conducted. From the viewpoint of student motivation, this article clarifies the influence of the class implementation method on the typical online classes currently being conducted, by a questionnaire survey after the experiment.

View full abstract

The COVID-19 has resulted in schools shut all across the world. As a result, education has changed to online platforms remotely from classical forms of learning. However, for many students, one of the most challenging aspect of online learning is that it is difficult to focus on the screen for long periods of time. Thus, students are easily distracted by social media or other sites during class. Therefore, it is necessary for the teachers to keep checking the students whether they are focused or not during class. In this paper, we propose a new method to surveil how students are concentrating during online class using eye tracker device. The experiment results showed that the proposed method is effective to evaluate concentration level for student in online learning.

View full abstract

This article introduces development of a media-processing technology for improving the accuracy of depth maps by combining depth estimation with image segmentation, a system that uses this technology to generate 3D images, and the experimental results using the images that generated by our system.

View full abstract

Historical background of current metadata managed by properties and values represented by Dublin Core has been described. Then distributed representation of natural language, Word2Vec, BERT of recent deep learning technologies has been explained, accompanied by the relationship with the metadata. Furthermore, the impact of the above deep learning technologies on the humanities field has been pointed out. Finally the importance of human resource development in related fields has been described.

View full abstract

From the viewpoint of an information researcher in a historical museum, two applications of metadata technology in our museum are introduced. One is the transition of the usage of metadata in the integrated retrieval system nihuINT, and the other is design of a conceptual model for the description of copied materials.

View full abstract

The possibility of an artist-in-residence program that supports creative activities directly related to a place will be examined from the perspective of international exchange and area studies, introducing specific examples from the standpoint of a practitioner.

View full abstract

Compressive spectral imaging with spectrally-varying PSFs is a technique for efficiently acquiring spectral images by a simple device configuration. This method requires a reconstruction process because it is based on compressed sensing. The spectral image reconstruction has challenges in terms of both speed and accuracy. In this paper, we propose a multi-staged neural network model, which is inspired by ADMM, to achieve fast and accurate reconstruction.

View full abstract

In this report, we shall show some results characteristics of the reconstructed holographic moving pictures employed blue-violet color laser light. By the multiplex holographic reconstructing process using blue violet color laser, we obtained nice moving pictures with high contrast.

View full abstract

We have jointly developed a road condition reporting system, called “Michirepo,” with Hiratsuka City, Kanagawa Prefecture. It has been operating as the official system of Hiratsuka City since March 2019. This paper reports on the latest version of Michirepo and the system’s operation for about two years since its launch.

View full abstract

What was the purpose of developing the “UD Digi Kyokasyo Type” that comes as standard in Windows 10? In this presentation, I will explain the design concept and the consideration given to reading and writing with the font shape, with reference to the reading and writing difficulties encountered by children in educational settings. We will also introduce the evidence for “UD Digi Kyokasyo Type”, the Latin font series and the stroke order fonts in our lineup.

View full abstract

In educational situations, it is essential to access printed matter such as textbooks, teaching materials, and study-aid books. However, it may be difficult for some students with disabilities to access printed matter such as textbooks. In this report, I will introduce the background and the current status of the efforts to make accessible textbooks and teaching materials. In addition, I will introduce the development of a digital large print textbook system and an accessible textbook viewing application “UD browser”.

View full abstract

ISO/IEC JTC 1/SC 29 is a sub-committee which has developed fundamental international standards such as JPEG and MPEG. Following the establishment of JPEG as a group, another group MPEG was established in 1988 before the establishment of SC 29. 32 years later, SC 29 has been significantly reorganized in 2020. Major changes are that the sub-groups under MPEG has been elevated to WGs. The activities as JPEG and MPEG continue. Here the brief history of SC 29 and the intent of its reorganization including its future are described. Some of recent activities in MPEG which consists of multiple WG/AGs are introduced as well.

View full abstract

Basic study is done for multi-symbol source coding using multi-alphabet arithmetic codes. We assumed the source model in which the probability of N-th probable symbol is given as R（0＜R≦1）times of that of (N-1)-th probable symbol. The arithmetic code register length is assumed to be 14 bit or more for 256 symbols. The assigned widths of the symbols other than the most probable symbol are fixed for variable valid area, and the rest of the valid area is assigned to the most probable symbol. The value of the valid area, at which the assigned widths of the symbols follow the information model, is defined in accordance with R, for the optimization of the coding efficiency, and the design example for two-divided valid areas is also shown. We pointed out the general learning issues for the adaptive coding.

View full abstract

Varroa mites parasitism of honey bees, one of the causes of bee shortage, is a daily problem for beekeepers. Since Varroa mites are very small (1-2mm) and visual measurement is not easy, there is a need to reduce the cost of measuring the parasitism rate of Varroa mites. In this study, we are developing an automatic counting method using image processing to count the number of mites that have fallen on the paper sheet laid on the bottom of the hive due to the application of disinfectant. In this paper, we implemented the proposed method as a smart phone application and identified issues that need to be solved for practical use.

View full abstract