Hiroaki Sawano

Road extraction by snake with inertia and differential features

We propose an extraction technique based on an active contour model (snake) considering inertia and differential features of an object in movie. In many methods for tracking a moving object, a snake is applied to a scene frame by frame, and initial positions of control points in a frame refer to results in the previous frame. We focus on inertia which works between previous and present frames. In this paper inertia is the tendency of a control point to resist changes in its state of motion in an image space. We propose an internal energy for snakes based on inertia of control points. Proposed method is applied to extract road geometry from a video camera equipped on the head of a vehicle. Internal energy functions based on differential features of road geometry is also introduced. Experimental results indicate the availability of the proposed method.

A Study on Scalable Object Replication Method for the Distributed Cooperative Storage System

Recently, a Peer-to-Peer (P2P) technology is applied for various systems because it enables to construct a system with high scalability and robustness.We have proposed a distributed cooperative storage system that can construct huge storage space by providing it on a P2P network.The storage system is based on an overlay network construction, and its hash space is created in an application layer.Each peer manages the certain range of the hash space, and an object is stored to suitable peers by using a hash value from the object.In the general P2P network, objects are replicated over many peers to achieve high availability.Some replicated methods have been proposed: objects are randomly assigned, assigned on a route path, or assigned to peers that the number of them is fixed by a network neighborhood depending on a distributed hash table.However, they are not so suitable since the scale of the network is not considered in these methods.Therefore, we propose distributed interval trees for setting the number suitably, and the availability is described with simulation.

A car-navigation system based on augmented reality

In this sketch we propose a display method for car-navigation systems based on an Augmented Reality (AR) technique. Carnavigation systems, as shown in Fig. 1, based on three-dimensional Computer Graphics (3D-CG) techniques with a digital map and GPS are widely used. However, if the digital map is not updated, it is very difficult for the user or the driver to understand and cognize that the display in Fig. 1 corresponds to the actual scenery mentioned in Fig. 2. In our approach, the system collects road geometry from a real-time video taken by an equipped camera on the front of the vehicle. The 3D-CG data is produced from the road geometry and is overlaid to the scene image as shown in Fig. 3. There are two major advantages in our method. One is to improve instantaneous visibility since the display corresponds to the scenery the driver watches. The other one is that it’s not required to prepare the detailed 3D-CG data in prior since the system collects the geometry in real time. The availability of the proposed method is indicated with a practical experiment.

Study on Multi-path Transmission Model Based on P2P Overlay Networks for Streaming Data

In the past years, the ALM (Application Layer Multicast) has led the rapid development of the real-time streaming data transmission technology. However, as focusing on the global performance and control, the individual efficiency in the ALM structure could not be guaranteed. Furthermore, according to the complex environment of the existing networks, the available bandwidth is often made full use of. Because of these, we started this research focusing on the single efficiency, targeting a reliable end-to-end streaming data transmission model to enhance the using of the bandwidth. Keeping the virtue of high scalability and reliability, we still consider applying the P2P overlay technology as ALM. The basic idea of our research is utilizing the transmission ability of free peers in the overlay network, constructing the multiple transmitting paths from the sender to the receiver. Our multiple paths method is enabled by the overlay P2P network structure called a management network. Besides, the robustness and recovery cost are also considered in our evaluation.

Flat shading road versus photo realistic road for AR-based car navigation

To guide a driver to his/her destination smoothly, there are carnavigation systems utilizing a driver’s view display as indicated in Fig. 1 (a). The authors have developed a car navigation system based on an augmented reality (AR) display method as shown in Fig. 1 (b)[2005] to use a real-time video from a camera equipped on the front of a vehicle. The abailability of the system has already studied in [2005]. Here the authors focus on the three-dimensional computer graphics (3D-CG) road by flat shading in our system. In this paper we propose a photo realistic road with velocity synchronization instead of the flat and static shading one.

A songwriting support system with a card

An user generated content is promoted by amateur since computer music software is improve as Hatsune Miku[CRYPTON FUTURE MEDIA ]. When lyrics and melody are created, rhyming words and the song may be considered. However, the experience and knowledge are required, and it is difficult for amateur to create songwriting. Therefore, we propose a songwriting support system with a card writing a word and a melody line by a user, and the target of the user is beginner of songwriting. A user writes a word or phrase and draws a curve for melody in a card. The system recognizes a curve shape by a computer vision (CV) technique, and the melody is created automatically in a prototype system. In this paper, the experimental result will be shown by the system.

Comparison of orientation-aware RSS-based localization algorithms

A large number of researches have been focused on indoor location tracking system over the past years. There are theoretical and empirical RSS-based algorithms proposed and many studies have been made by comparing the performance of these algorithms. However, there is a lack of theoretical basis and understanding on the orientation of the user. This paper presents a model for an orientation-aware indoor location tracking system using wireless sensors based on IEEE 802.15.4 standard protocol. For path loss model, our experiment shows a better prediction of RSS when attenuation factors for different orientations are considered in the model. On the other hand, orientation-aware fingerprint algorithm, which currently stands as one of the most accurate RSS-based localization algorithm, is also examined in our experiment to have a basis of comparison with an empirical algorithm.

Analysis of Diverse Tourist Information Distributed Across the Internet

Herein, we propose and discuss a new method for analyzing various types of tourist information about the Suwa area of Nagano Prefecture, Japan, available on the Internet. This information includes not only long sentences that can be found on web pages and in blogs, but also short sentences comprising a few words posted on social media. In this paper, we propose a novel method based on a neural network, called paragraph vector, for expressing relationships between words included in sentences. Our method achieves high retrieval accuracy even across social media posts comprising just a few words. Based on our evaluation results, the proposed method outperforms the conventional information retrieval technique wherein sufficient accuracy cannot be achieved as it is based on the occurrence probability of words in sentences. This improvement is achieved by using the word order as an input feature to the neural network model.

A Community-based P2P OSNs using broadcast encryption supporting cross-platform with high-security

Online Social Networks (OSNs) are flourishing such as Facebook and Google+, where people share recent status and personal files with friends. However, security and privacy have become major concerns in existing C/S based OSNs. In this paper, we propose a novel scheme called a Community-based P2P OSNs with Security Enhancement (ComPOSE) that supports cross-platform availability and reputation mechanism in stringent privacy requirements. For the first time, we formulate the privacy issue and design goals for P2P based OSNs. We firstly introduce broadcast encryption into P2P community-based social networks together with hierarchical encryption algorithms in order to decrease the system overhead. In ComPOSE, we employ the Open Directory Project to classify and encode groups. The RESTful web-services help to ensure cross-platform availability and transmission security. As a result, thorough security analysis and performance evaluation on experiments demonstrate that the ComPOSE scheme indeed accords with our proposed design goals.

AKaTool: An association chart production tool for nursing education

We propose an association chart production tool called AKaTool. A nursing student learns the association chart to solve a nursing problem logically. In ordinary education, a creation method of the chart is by handwriting, and it spends not only consideration time but also a lot of writing and editing one. AKaTool is used to reduce the chart writing and editing costs for the nursing student. In contrast, a lecturer spends a time to correct the chart written by the student. We implement a highlight function that indicates a path from patient information to nursing diagnosis for lecturer correction support. In this paper, we introduce AKaTool proposed by us, and it is discussed in the experimental result.