The purpose of this study is to automatically detect abnormal driving behavior from in-vehicle camera video. The previous method used a Multi-stream CNN based on the original image and optical flow. However, this dataset could not outperform the accuracy rate of a CNN using the original image as input. Therefore, we propose a method to improve the accuracy of Abnormal driving behavior detection by using ST-GCN with skeleton as input and combining it with CNN using the original image as input. Furthermore, we prepared two input coordinate systems (Cartesian and polar coordinates) and four data augmentations (affine transformation, left-right flipping, dropout for joints, and adding Gaussian noise) for ST-GCN. we investigated combinations of ST-GCN input coordinate systems and data augmentations that are effective for this task.

In recent years, more and more server administration exercises have been conducted at universities. Some students get stuck on the answer to a question. In this case, the student requests a hint for the question. The teacher hears the student, assesses the progress, and offers a hint. Also, teachers want students to conduct trial and error on a problem before asking for hints. Many students ask for hints during exercises, which is a burden for teachers. We develop a system that presents hints based on the student's learning progress. The teacher registers the learning progress and hints in the system in advance. The student's command input/output and edited files are monitored by the system to check the student's progress. The system can suggest many hints based on learning progress. As a result of the evaluation experiment, unnecessary hints were not presented to the students. Also, the system can guide students to conduct trial and error in the learning situation the teacher expects. In this paper, we describe the developed system and the results of evaluation experiments using the system.

In a social context where digital signage and video streaming have become widely used, it is necessary to clarify the behavior of visual search for moving visual targets. This study focuses on an inconspicuous visual component, spatial frequency, and examines the relationship between the moving speed of the target in a visual search task and the threshold of spatial frequency that determines whether the target can be searched or not. In the experiment, scan paths were evaluated based on the distance between gazing points and the target image as the gaze error, and also machine learning. The results showed that the threshold value of b as an index of spatial frequency tended to increase from 0.3 to 0.6 as the moving speed of the visual target image increased from 0.2 to 3 times /s. Thus, these results quantitatively clarified the characteristics of the threshold of spatial frequency, and the change of the threshold with the change of movement speed. Furthermore, it was found the gaze error can be used to determine the threshold of the β value.

In this paper, we propose a method to detect driving scenes where cognitive function can be evaluated. This method defines assessable scenes as those composed of three elements: road structure, appearing objects, and operations. It detects scenes composed of these three elements, which are arbitrarily set. When detecting targets composed of multiple information sources, and for targets where the pre-description of useful feature vectors is difficult, multimodal deep learning is used. While there are cases where an intermediate fusion model structure is used in existing research, it has been suggested that such models face challenges with hyperparameter tuning and may fail to learn the inter-modality relationships when there are discrepancies in the amount of information each modality provides. Therefore, in this paper, a new model structure that incorporates an attention mechanism into a late fusion model is proposed. This model not only enables individual evaluation of each modality constituting the scenes and achieves the final detection result, but also provides a structure with high readability regarding how the detection results are produced. In experiments, this method is compared in terms of detection accuracy with the intermediate fusion model structure used in existing research, and improvements in both recall and precision were confirmed.

Predictive maintenance is a technique to perform maintenance before failures happen by finding their Indications in advance and is a key to streamlining Maintenance operations and reducing the downtime. Methods for predictive maintenance based on anomaly detection using deep learning have been actively studied, but the identification of anomalous sensors remains a challenging task. As sensors corresponding to the cause of anomaly do not necessarily indicate large anomaly scores, it is important to watch how a model computes the scores. In this work, we use a graph neural network for anomaly detection and isolation. The vertices of the graph that appears in the network correspond to the sensors, so we can interpret the relevant weights as the relationship between the sensors. We specifically used a sparse variant of graph attention network for anomaly detection and isolation. We applied it to real-world storage battery data and confirmed the effectiveness of the method.

Abridgement is a form of summarisation task that offers an innovative learning approach for enhancing reading comprehension within language education. To scale up the application of Abridgement Method at Kobe Tokiwa University, we propose the novel Character-Pyramid (CP) method and a semi-automatic abridgement generation methodology. The CP method is designed to automate the evaluation of abridged summaries and is accompanied by the semi-automatic abridgement method. Our data analysis reveals that the CP method diverges manual assessment by an expert, whereas the semi-automatic abridgement method demonstrates its reliability.

Smart cities, which can monitor the real world and provide smart services in a variety of fields, have improved people's living standards as urbanization has accelerated. However, there are security and privacy concerns because smart city applications collect large amounts of privacy-sensitive information from people and their social circles. Anonymization, which generalizes data and reduces data uniqueness, is an important step in preserving the privacy of sensitive information. However, anonymization methods frequently require large datasets and rely on untrusted third parties to collect and manage data, particularly in a cloud environment. In this case, private data leakage remains a critical issue, discouraging users from sharing their data and impeding the advancement of smart city services. This problem can be solved if the computational entity performs anonymization without obtaining the original plain text. This study proposed a hierarchical k-anonymization framework using homomorphic encryption and secret sharing composed of two types of domains. Different computing methods are selected flexibly, and two domains are connected hierarchically to obtain higher-level anonymization results efficiently. The experimental results show that connecting two domains can accelerate the anonymization process, indicating that the proposed secure hierarchical architecture is practical and efficient.

In recent years, heavy snowstorms have caused traffic disruptions, gridlocks, and social problems. One of the reasons for this is that vehicles get stuck in snowdrifts and snow accumulations caused by snowstorms. To prevent this, it is necessary to know the height of snow accumulation even in poor visibility conditions, to stop and to reverse safely. In this study, we develop a sensor to determine the height of snow on the road even in poor visibility conditions, and conduct an experiment to determine the height of snow accumulation under natural snowfall conditions.

Substrate alignment is an important topic in the micro fabrication technology to guarantee the development of secure micro-structured devices. There often exists an alignment error due to operator habits in the alignment process. Here, we modeled such error with the normal distribution, and the suggested model was validated by means of QQ plots and Shapiro-Wilk test. In addition, we pointed that the error in the alignment process would be reduced via calibration with statistically extracted properties of the error model. Our investigation therefore contributes to secure devices developed with the micro fabrication technology.

As an omnidirectional sound source tracking using a square arrangement microphone array, a role selecting method has been developed. In the method, a role selection of each microphone pair is carried out to classify the estimated time-difference-of-arrivals to two roles based on a sensitivity difference between microphone pairs. However, there exist directions that two pairs have a high sensitivity at the same time. This situation prevents an assumption for applying the method. In this paper, square edge pairs also are incorporated to candidates for avoiding such a situation and aiming a performance improvement. The effectiveness of the method is shown through several experimental results.