In recent years, autonomous mobile robots, such as self-driving robots, have attracted much attention, and their practical application is expected. A range sensor, which collects information about the external world around a mobile object, is one of the most important devices for safe and accurate control of an autonomous mobile robot. It is not uncommon for autonomous mobile robots to become dirty during operation in outdoors. In particular, since the stains of the sensor affects its control, an automatic detection system is necessary. In this study, we proposed a 3D point cloud processing method (Maximum Value Collection) that detects the stained area on the optical sensor using the maximum value of the lightreceiving intensity values, and conducted a demonstration experiment applying the method to a 3D laser scanning system in optical. Our experimental results show that the proposed method is useful for estimating the stains on the optical sensor surface and its location.

Holography records the amplitude and phase of the light wave from the object to be recorded as interference fringes by interfering with the reference light. Then, it is a technology to reproduce the complex amplitude distribution of light waves by using the diffraction phenomenon. Computer-Generated Holograms (CGH) has made it easier to realize interference fringes. In holographic memory, the lens term is used for image multiplexing / demultiplexing, and research is being conducted to improve the deterioration of the reconstructed image by image processing. This paper evaluates the creation of CGH, image multiplexing / demultiplexing by lens terms, and noise removal by filters as digital technologies by simulation. In addition, the limit of the total number of page data is shown, the improvement of the reconstructed image by image processing is confirmed.

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(Single Shot MultiBox Detector) target detection algorithm. In the initial stage of the experiment, we increase the number of data and select a convolutional neural network structure with high feature expression ability. The accuracy of the results was improved by 10% compared to the results at the beginning of the experiment.

Sequential photography is widely used for sports visualizations because it illustrates one motion with a single static image. However, because sequential photographs are usually displayed on paper or monitor and their sizes are smaller than life-size, observers cannot easily recognize the actual size of the motion. In this paper, we propose a life-size visualization method of sequential photography by using mixed reality (MR) technology. With our method, the observer wears a video-see-through head-mounted display and captures a motion of a player by using a hand-held camera. Our method synthesizes sequential photography from the video and visualizes it in life-size in MR space. To confirm the effectiveness of the proposed method, we conducted a user study where we compared a smaller visualization with a monitor display, a life-size visualization with a projector, and a life-sized visualization with our method. The results suggest that the life-size visualizations with a projector and our method allow observers to recognize scales of motions accurately and confidently.

There have been proposed methods to depict impossible figures, which look like impossible objects by using optical illusion objects. The optical illusion objects are 3D objects that look like impossible objects when viewed from a specific viewpoint. There have been proposed several methods to animate impossible figures by using 3DCG models of optical illusion objects, which are remodeled according to the viewpoint. However, these methods are dedicated to depicting only impossible figures, not to depicting ordinal 3DCG models. We propose an illusion representation model, independent from rendering environments, consisting of both an optical illusion object given as a 3DCG model and an algorithm to dynamically remodel the 3DCG model according to the viewpoint. The illusion representation model can be processed to match the depth of the impression recalled from the remodeled optical illusion objects with the depth of the impossible objects without any side effects on the rendering pipeline. Thus, it can co-operate with ordinal 3DCG models as well as the illusion representation models. Furthermore, we analyze the properties of the shadows of impossible figures and propose a shadowing method based on this analysis.

To depict liquid realistically in computer graphics (CG) works, it is necessary to represent bubbles in the liquid. In conventional studies, liquid and gas particles have been utilized to represent bubbles in a liquid. On the other hand, the gas particles increase computational complexity. Additionally, the conventional fluid simulations can not utilize an implicit method to calculate the flow of liquid with high viscosity. This study proposes a simulation method to represent bubbles in liquid with various viscosity without gas particles. To calculate bubbles, we utilize the conservation of the mass of bubbles. For this purpose, we apply pressure calculated from the bubble density as the pressure boundary condition at the bubble surface to the Navier-Stokes equation. By assuming that the sum of vertical velocity to the surface at the bubble is zero, the volume change of the bubble can be minimized. The simulation method does not require gas particles because the two conditions conserve the mass and the volume of a gas. The proposed model can be incorporated into the material point method, which can represent liquid with various viscosity in a short time. Owing to the mentioned characteristics, the proposed model can realistically represent bubbles in a liquid such as water or honey and contribute to improving the quality of CG works.

The number of pancreatic cancer patients in Japan has been increasing. There is a shortage of pathologists who can diagnose pancreatic cancer by microscopic observation of tissues and cells, and the development of automatic diagnosis systems based on digitalization and machine learning methods is in progress. However, it has been pointed out that machine learning using ordinary RGB images has poor interpretability of classification models. In this study, we analyze pancreatic cancer cytology specimens captured by a hyperspectral camera using supervised and unsupervised learning in machine learning and perform spectral interpretation of the models. As a result, the t-SNE algorithm shows excellent visualization performance in unsupervised learning, and the LightGBM algorithm achieves 93% accuracy in discriminating between benign and malignant classifications and 82% accuracy in discriminating between malignant and benign classifications in supervised learning. The model also suggests that two main wavelength ranges are important for diagnosis. Although further research on the accuracy and the number of cases is needed to introduce the hyperspectral camera into the medical field, this study suggests that the hyperspectral camera has a high potential for application in cytology.