Evacuation training is indispensable in terms of surviving disasters. Recently, virtual reality technologies have been used to facilitate evacuation training. The authors previously proposed the Evacuation Training in Metaverse (ETM) system. This system is currently being evaluated in a prototyping phase, where many users can spend time in a three-dimensional virtual university campus. In the ETM system, an earthquake occurs suddenly, and the users must determine various actions, e.g., whether to evacuate and where to evacuate. However, the ETM system does not always have a sufficient number of active users. Thus, the ETM system can generate non-player characters (NPC) based on the evacuation logs of previous participants. This paper overviews the ETM system and describes the method used to generate NPCs. In addition, the results of a preliminary comparative experiment are discussed.

Evacuation training should be more realistic. We aim to develop a VR-based evacuation training system that includes gaming elements but completely separates evacuation training and game. In addition, the training system should simulate earthquakes that suddenly occur. Before starting the development, we got interest ed in observing the evacuation behaviours of participants in a VR earthquake simulator. Thus, we conducted a preliminary comparative experiment that examined the differences between one-participant and two-participant settings in the simulator. As a result, participants tended to keep moving in the one-participant setting and consequently completed evacuation faster. However, participants tended to have more discussion and synchronisation in the two-participant setting, resulting in a slower evacuation. These findings have the potential to be beneficial to development.

Evacuation training is crucial for protecting human lives from natural and man-made disasters, but it should be more realistic to achieve training effects. In this study, we focused on expressing disaster situations using markerless augmented reality to achieve realistic evacuation training. We prototyped a scenario-based evacuation training system that superimposed three-dimensional computer graphics of disaster situations (e.g. fire and debris) onto real-time vision (captured by Android tablets or smartphones) using ARCore and Unity3D. Through preliminary experiments, we found that the prototype system can provide realistic expression and potentially be used for evacuation training, but we have not yet clarified the training results and how the expressions influenced participants’ emotions.

In Japan, a disaster-prone country, foreigners may fail to survive in disasters because they lack disaster-related knowledge, so disaster education for foreigners living in Japan is necessary. However, many foreigners are unwilling to spend time on disaster education. To improve this situation, gamification is employed. Gamification is to use game design elements or mechanisms in non-game contexts. For example, in the educational field, gamification makes learning easy and attractive, which motivates participants and encourages engagement in learning activities. This paper explores an information and communication technology–based system with application of gamification in disaster education for foreigners living in Japan.

This study develops a system that focuses on proper hand-washing patterns to examine whether hands are being washed properly. By comparing images of hand-washing taken at sinks and the like with model images, it informs users in real-time whether they are using proper hand-washing patterns. Optical flow and skin-color pixel areas are used as feature amounts for recognizing proper hand-washing patterns. SVM (Support Vector Machine) is introduced as the recognition model in this system.

Evacuation training is an important component of disaster education and survival. Evacuation training using a virtual reality (VR)-based disaster simulator that provides a highly immersive simulated evacuation experience (SEE) has attracted significant attention. To improve the training effect, we propose a failure-enhanced evacuation training model based on Kolb’s experiential learning theory. Our model aims to purposefully induce participants to succumb to conformity bias and fail to evacuate during the first SEE because inactive evacuees (i.e., people who are not evacuating speedily or not starting their evacuation) are simulated in a VR-based disaster simulator. The participants are expected to overcome failure in the second SEE via reflection and conceptualization. A preliminary comparative experiment focused on how simulated evacuees influence the SEE of participants in a VR-based disaster simulator. Results indicated that failure-enhanced evacuation training can successfully improve the training effect.

Since conventional evacuation drills do not adequately simulate disaster situations, participants do not feel a sense of tension during evacuation. We developed a game-based evacuation drill (GBED) system that focuses on situational and audio-visual realities and scenario-based interactivity. To improve the visual reality in a GBED, we adopt simple augmented reality (AR) and a binocular opaque head-mounted display (HMD). The simple AR represents vague extensive disaster situations (i.e., rain, fog, smoke and fire) by superimposing the overall disaster situations (dynamic three-dimensional computer graphics) onto the real-time vision captured by a stereo camera (attached to the HMD).

Disaster education focusing on how we should take immediate actions after disasters strike is essential to protect our lives. We proposed Immediate-Action Commanders (IACers) training and prototyped three IACers training systems using augmented reality (AR) and head-mounted displays (HMDs). The systems superimpose interactive virtual objects onto HMDs’ real-time vision or a trainee’s actual view.

This paper presents an approach for improving the use of VP-tree in video indexing and search. VP-tree is one of the metric space-based indexing methods used in multimedia database search and data retrieval. The proposed approach focuses on the trigonometric inequality for compressing the search range, thus, improving the search performance. A test result of using 10, 000 video files shows that this method reduced search time by 5 - 12%, compared to the existing method using the AESA (Approximating and Eliminating Search Algorithm) algorithm.

In this paper, we describe a Japanese speech corpus collected for investigating the speech variability of a specific speaker over short and long time periods and then report the variability of speech recognition performance over short and long time periods. Although speakers use a speaker-dependent speech recognition system, it is known that speech recognition performance varies pending when the utterance was uttered. This is because speech quality varies by occasion even if the speaker and utterance remain constant. However, the relationships between intra-speaker speech variability and speech recognition performance are not clear. Hence, we have been collecting speech data to investigate these relationships since November 2002. In this paper, we introduce our speech corpus and report speech recognition experiments using our corpus. Experimental results show that the variability of recognition performance over different days is larger than variability of recognition performance within a day.

This paper reports an evaluation of European Telecommunications Standards Institute (ETSI) standard Distributed Speech Recognition (DSR) front-end through continuous word recognition on a Japanese speech corpus and proposes a method, the Bias Removal Method (BRM), that reduces the distortion between feature vector and VQ codebook. Experimental results show that using nonquantized features in acoustic model training procedure can improve the recognition performance of DSR fornt-end features and that the proposed method can improve recognition performances of DSR front-end feature.

The final purpose of the authors is to achieve indicative summarization for retrieved text. The paper examines the effectiveness of a method of automatically deriving Japanese compound keywords based on rules of dependency relationship and restrictions on them. Then, an importance measure of derived keywords is discussed. The derivation method can derive more significant Japanese compound keywords that do not appear in the original text.

The vector space model (VSM) is a conventional information retrieval model, which represents a document collection by a term-by-document matrix. Since term-by-document matrices are usually high-dimensional and sparse, they are susceptible to noise and are also difficult to capture the underlying semantic structure. Additionally, the storage and processing of such matrices places great demands on computing resources. Dimensionality reduction is a way to overcome these problems. Principal component analysis (PCA) and singular value decomposition (SVD) are popular techniques for dimensionality reduction based on matrix decomposition, however they contain both positive and negative values in the decomposed matrices. In the work described here, we use non-negative matrix factorization (NMF) for dimensionality reduction of the vector space model. Since matrices decomposed by NMF only contain non-negative values, the original data are represented by only additive, not subtractive, combinations of the basis vectors. This characteristic of parts-based representation is appealing because it reflects the intuitive notion of combining parts to form a whole. Also NMF computation is based on the simple iterative algorithm, it is therefore advantageous for applications involving large matrices. Using the MEDLINE collection, we experimentally showed that NMF offers great improvement over the vector space model.

In the current natural language interface system, it is impossible to understand erroneous sentences. In order to realize the superior one, the automatic error recovery for erroneous sentences is one of the problems to be solved. The method to apply the LR parsing strategies is one of the famous approaches, however it takes many time to parse the sentence. This paper shows the method to improve the time efficiency keeping the accuracy of the traditional method.

A trie structure is frequently used for various applications, such as natural language dictionaries, database systems and compilers. However, the total number of states of a trie (and transitions between them) becomes large, so that the space cost may not be acceptable for a huge key set. In order to resolve this disadvantage, this paper presents a new scheme, called a "two-trie", that enables us to perform efficient retrievals, insertions and deletions for the key sets. The essential idea is to construct two tries for both front and rear compressions of keys, which is similar to a DAWG (directed acyclic word-graph). The approach differs from a DAWG in that the two-trie approach presented can uniquely determine information corresponding to keys while a DAWG cannot. For an efficient implementation of the two-trie, two types of data structures are introduced. Theoretical and experimental observations show that the method presented is more practical than existing ones considering the use of dynamic key sets, information storage of keys and compression of transitions.