To acquire geometrical relationship between the flying robot and its surroundings, LSD-SLAM（Large Scale Direct Monocular-Simultaneous Localization and Mapping）is frequently applied to the image acquired by the camera attached to the flying robot so that the map for the surroundings is obtained, but its problem is that 3D structure of object surfaces that do not have enough texture cannot be reconstructed. This paper studies a method that aims at reconstructing 3D structure of the surroundings by projecting a grid pattern from the laser projector attached to the flying robot. Specifically, if 4 x 4 grid pattern is projected, the projected pattern is geometrically deformed according to Homography transformation if the projected object surface is a plane. If the pattern is projected to discontinuity between two planes, the intensities of the projected grids’ positions predicted by the Homography transformation are analyzed so that whether the pattern is projected over the discontinuity is judged. Promising experimental results are obtained.
Animal identification method is required for the purpose of controlling animals. However, there is no effective technique to identify animals perfectly. The suggested method uses the visual information and movement information for identification. Here, extraction of the moving object is necessary in order to identity movement. So, we proposed an automatic tracking system using Snakes. This system extracts the moving object from moving picture.
Recently, we have used many videos by various image display devices. In particular, since an opportunity of viewing contents in a large size display or a mobile device has been increased, it is one of the important problems to prevent visually induced motion sickness (VIMS). Then, in this study, in order to decrease an adverse effect on a human body due to VIMS, a decision method of VIMS in consideration of high-speed processing is proposed. The histogram is calculated by using their vectors, which are obtained by coded data in a video content, or which are estimated by decoded data. The VIMS state is determined by using the degree of similarity to grasp the temporal change of the histogram. The effectiveness of the proposed method is considered by the simulation experiments. By the experimental results, it is revealed that the pseudo motions of shake affect the values of the degree of similarity and the simple proposed method can decide the state of VIMS.
Gait recognition, identifying a person by the way they walk, is an unobtrusive way of performing human recognition. The task can be complicated by the presence of covariate factors such as clothing, carrying condition, walking surface, elapsed time, etc. Amongst these factors, clothing is the most challenging one, as it may cover a significant amount of gait features and make human recognition difficult. Since the location of occlusions may differ for different clothing types, relevant gait features may become irrelevant when clothing-type changes, and the use of occluded gait features can hinder the recognition performance. To address this problem, we present a wrapper-based feature selection method that uses evolutionary computation to extract the most relevant and informative gait features for human identification. The proposed method is evaluated using OU-ISIR Treadmill dataset B and the experimental results indicate that our feature selection method significantly improve the performance of gait recognition.
With drastic increase in the collection of digitalized artworks, archiving the artworks database by human become extremely difficult. Hence, the aid of machine learning technique is important to automate such task. Recently, Convolutional Neural Network (CNN) has become a popular choice for feature extraction and classification. Many studies indicate that features learnt by CNN through ImageNet can be generalized to many other similar tasks. This paper demonstrates that such CNN features can even be generalized to fine-art paintings classifications. In this work, we focus on Style, Genre, and Artist classifications problems. We used the recently available large-scale Wikiart paintings dataset that is publicly available. We show that an end-to-end trained CNN is able to achieve descent performance for these tasks. More importantly, an ImageNet pre-trained CNN is able to achieve better recognition. In addition, fine-tuning allows the pretrained CNN to better adapt to the new tasks, resulting in significant improvement. Furthermore, we observe that Softmax Regression and Support Vector Machine performed on par in our experiments.
Currently, appearance of drones makes aerial image more popular insecurity and disaster response scene. Usually, the pictures taken from midair have low visibility, because of low light or shadow. Multiple Image noise reduction technique is effective way to improve visibility for high noisy image. But sometime we cannot obtain multiple image of same point of views because of camera motion. In this case, simply composing multi images generates a double image. We present a Noise reduction algorithm using composition of multi aspect images. This method adds pixels using addition of space direction (in frame) and time Direction (between frames). This algorithm prevents unnatural double images and reduces noise.
Nowadays, the efficient searching system and techniques which acquire the new knowledge correctly from the digital image database is needed. Using image texture feature including useful information is effective for understanding and describing the database. In this paper, the Variable Texture Feature Model using SOM of SOMs (SOMn) method which represent the overall database image feature information is proposed. The correct Variable Texture Feature Model is not only useful for the index to texture classification but also can create new images from known images. Experiment shows the Variable Texture Feature Model has the effectivity to supplement the texture feature for classification, and can select the image which represent the texture class automatically, create the new texture images from known images.
Users of smartphones and/or tablet terminals browse and download confidential document files routinely. Then, a higher security level is needed for smartphones and tablet terminals than conventional mobile phones. Therefore, in this paper, we propose an image-based user authentication method using swipe operation which is peculiar to touch screen terminals. The Security level of the proposed method is higher than 4 digit PIN authentication. Moreover, we conducted several experiments to evaluate the authentication time and the resistance against recording attacks. The results of these experiments clarified the usefulness of the proposed method.
We have studied a technique that can protect the copyrights of digital content for 3D printed products. It embeds information on copyrights inside real objects produced with 3D printer by forming fine cavities that cannot be observed from the outside. The disposition of the fine cavities expresses the information. We read out information from inside real objects non-destructively by analyzing the temperature profile of the surface of the objects using thermography. In this study, we evaluated the readability of embedded information relating with the structure parameters of the fine cavities and clarified the conditions for forming cavities inside the fabricated object.
The purpose of this study is visualization of linear algebra for educational use. Some students tend to learn how to solve problems when they learn mathematics. Then, they don't extend to the understanding of the essence of the operations. Therefore, we aim at the visualization promoting essential understanding about the operations of the linear algebra in particular. The visualization policies are as follows. First, 2×2 or 3×3 matrixes are regarded as 2D or 3D coordinate transformation. In case of the transformation, we visualize the lattice in plain or space. We decide presentation methods of each operation that it is easy for students to understand from the characteristics of the lattice before and after this transformation. Next, we work up each presentation methods into a application as a teaching material.
The purpose of this study is improved visualization of document reference relationship. It is important to survey the related studies. However, it is difficult to discover useful information because it is necessary to look over a huge number of documents one by one. In previous visualization method for document reference relationship proposed by Okada et al., it is difficult to grasp the flow of the whole. In this research, we improve visibility by modifying the node locations and link drawing. Also, we verify the degree of similarity.
This paper proposes a method that recognizes the grasping action of the upper-half of the human body from RGBD time-sequential images in case that an object is placed at different positions on a desk and that the object is grasped by the human in different manners. The proposed method consists of modules for extracting human body features and for recognizing human actions. In the feature extraction module, to Vemulapalli et al.’s twenty features based on relative poses between the links connecting human joints, five features concerning the neck as well as the thumbs and finger tips of both hands are newly added. In addition, a partial skeleton model that uses only the features in the upper-half of the body is utilized. The action recognition module uses a linear-SVM (Support Vector Machine) for each action to be recognized. Experiments for recognizing 10 actions that include nine combinations of three grasping manners and three different object positions as well as “do nothing” show promising recognition accuracies of the proposed method.
We implemented a prototype web system that presents recovery status of disaster-struck areas of the Great East Japan Earthquake in the previous research. However, there exist several problems for the system to be run. Therefore, in this study, we rebuilt the system to be available for actual operation. In the previous system, we first put together the pictures taken at our field work manually and then display them on a map. It is desirable that organizing pictures to be displayed on the map should be done automatically, because we have now more than 5,000 pictures and it is planed that incorporating pictures that are taken by someone other than our group members into our system. In this study, we implemented a system that first cluster pictures based on the location information (longitude and latitude information) automatically and then display them on the map. Moreover, we improved the user interface of the system.
Analysis of cellular behavior is significant for studying cell cycle and detecting anti-cancer drugs. It is a very difficult task for image processing to isolate individual cells in confocal microscopic images of non-stained live cell cultures. Because these images do not have adequate textural variations. Manual cell segmentation requires massive labor and is a time consuming process. This paper describes an automated cell segmentation method for localizing the cells of Chinese hamster ovary cell culture. Several kinds of high-dimensional feature descriptors, K-means clustering method and Chan-Vese Active contour Model are used to extract the cellular regions. The region extracted are used to classify phases in cell cycle. The segmentation results were experimentally assessed. As a result, the proposed method proved to be significant for cell isolation. In the evaluation experiments, we constructed a database of Chinese Hamster Ovary
Cell’s microscopic images which includes various photographing environments under the guidance of a biologist.
In recent year, the shortage of nurses is so serious that the actualization of RSN (Robotic Scrub Nurse), which can recognize surgeries’ stages and pass the necessary tools to the surgeons is strongly desired. This paper proposes a method that can recognize five stages of suture surgery: disinfection, anesthesia, washing, suture and tying, by recognizing surgical hand actions from video sequences. The proposed method consists of modules for detecting the starting points of each action and for recognizing actions. In the starting point detection, SVM (Support Vector Machine) is trained using video segments that include start points and is used for the detection. In the action recognition, neural network is trained using Sliced Bag-of-Words Sections(SBoWS) features of training video sequences, and is used for the recognition. Recognition experiments for extracted hand action video sequences and continuous video sequences show good recognition accuracies.
Towards the realization of the 3D reconstruction of the cerebral arteries from MRA images, this paper studies how velocities of blood flows influence on intensities of MRA images. Specifically, the following two studies are researched. First, in sections of blood vessels, blood flow velocity distributions obtained from MRA images and by a fluid dynamics based 3D simulation, which computes the blood flows based on the 3D structure of blood vessels reconstructed from thresholded MRA images, are compared. Experimental results indicate that intensities inside blood vessels are affected by blood flow velocities. Second, whether binarized local areas that include blood vessels correspond to inside and outside the blood vessels is studied. Experimental results show that binarization results using the threshold obtained by Otsu’s method could discriminate the inside (blood vessels) and outside (surrounding tissues).
In this paper, we propose a method to extract nuclear area in pathological images for diagnosis of endometrial endometrioid adenocarcinoma. Firstly, we divide on image into small areas called superpixels. Secondly, we classify the superpixels into two classes (nucleus or background). Finally, we specify seed points using previous results, and perform division by GrabCut. By the specification of the seeds from the result of GrabCut, we have made it possible to perform highly accurate extraction.
In this paper, we propose a motion interpolation method using the parameters based on adjectives. We have conducted a questionnaire experiment where subjects are asked to annotate ten example walking motions using 41 pairs of adjectives. Based on the results, we selected 27 pairs of adjectives that are effective for motion parameterization and determined four primary components by categorizing the adjectives. Our motion interpolation method allows the user to create various styles of motions through the four primary parameters and any combinations of additional adjectives from 27 pairs of them. We present the results of our experiments and demonstrate the advantage of our method.
Users of smartphones and/or tablet computers browse and download confidential document files routinely. Therefore, the higher security level is needed for smartphones and tablet computers than conventional mobile phones (feature phones). From this kind of background, Kosugi et al. proposed SWIPASS, an image-based user authentication method for touch screen devices by using images shot by user oneself and gotten from the WEB. SWIPASS has resistance to observation attack, recording attack, and smudge attack, the most serious threats for touch screen devices. However, there is an issue such that when there exist images similar to the pass image in the terminal, legitimate users do not authenticate properly because they cannot recognize the pass image accurately. Then, in this paper, we propose a method to eliminate images whose likelihood of being recognized as the pass image wrongly are high from candidates to be displayed as a decoy image. We moreover conduct several experiments to verify the effectiveness of the proposed method.
Visually impaired persons can use screen reader functions to operate application software on smartphones. However, concentration to operation is needed for accurate operations, or time to wait for speech annotations is needed. In case of music player application, small and complex menus and buttons cause difficulty for visually impaired persons’ uses. Thus, a prototype music player application on Android based smartphones are implemented for visually impaired persons, by replacing menus and buttons with tap and flick operations.
The purpose of this study is to propose a visualizing method for comparing multiple stock data. Though comparing multiple stock data is useful for stock trading, exiting tools are unsuitable for comparison because they prioritize analyzing one brand. Exiting visualizing methods for large scale time-series data are unsuitable for stock data because trend of price’s change is more important than amount of price in analyzing stock data. In this study, we propose visualizing method by combining line graph, Treemap and heat map.