A non-photorealistic rendering (NPR) method for automatically generating checkered pattern images from gray-scale photographic images using Prewitt filter with an expanded window size has been proposed. In this paper, we propose an extension of the conventional method to apply to RGB-D images. Our method can change the size of the checkered patterns depending on the depth. To verify the effectiveness of our method, we conducted experiments that are visually confirmed the checkered patterns by changing the parameters in our method. As a result of the experiment, it was found that the size of the checkered patterns can be automatically changed according to the depth. Additionally, it was found that the size and density of the checkered patterns can be adjusted by changing the value of the parameters.

抄録全体を表示

With the continuous improvement of social development level, traffic has become complicated. Therefore, the detection of pedestrian and vehicles becomes important. There are many application scenarios for pedestrian-vehicle detection, such as autonomous driving and transportation. This paper mainly introduces the research status of pedestrian-vehicle detection, analyzes the advantages and disadvantages of various current target detection algorithms, and focuses on YOLOv5 algorithm. Because the YOLOv5 model is much smaller than YOLOv4, and YOLOv5 also has strong detection ability. Finally, YOLOv5 is used to carry out pedestrian-vehicle detection experiments. The results the detection accuracy is improved slightly.

抄録全体を表示

Human experts have assured social infrastructure inspections, and recently an automated inspection is expected as an integrated system of the flight vehicle with software algorithms of the image processing. For the submillimeter-width concrete-crack detection, we have introduced Morphological Component Analysis (MCA) to be able to find those positions. Conventional image processing methods work well for thick-width crack detections, while the thin-width crack detection is highly difficult. We successfully demonstrated a concrete crack detection from images through proximity cameras in a specialized multi-copter and MCA-based crack position estimations and the linear regression-based sub-pixel width estimation were integrated together. It will open a new door to engineering in the actual field work, which can be widely applicable for social infrastructure inspections in general.

抄録全体を表示

This paper focuses on the R&D direction and the business strategy of EV firms and battery makers with reference to Porter’s productive frontier. M. E. Porter (1996) claimed that the productivity frontier represents the maximum value that the organization can deliver at any a given cost, using technologies, skills and purchased inputs. He argued that strategic decisions are ones that are aimed at differentiating an organization from its competitors in a sustainable way in the future. We use the patent information of EV firms (Toyota, Tesla, Volkswagen) and battery makers (Panasonic, CATL, LG Chem) as the cases. We examine our propositions by social network analysis and text mining. The analysis in this paper includes: 1) trying to distinguish between differentiation and cost leadership strategy from R&D direction, and visualizing productivity frontier, 2) making discussing on the inter-organizational relation of EV firms and battery makers. In this paper, we clarify those patterns of cooperation EV firms and battery makers.

抄録全体を表示

Writing VDM++ specifications is difficult. The existing method can automatically generate only type and constant definitions in VDM++ specification from natural language specification by using machine learning. This paper proposes a method to generate classes and instance variable definitions in VDM++ specification from natural language specification to improve the usefulness of the existing method. From the evaluation experiment by using F- values, it has been confirmed that the proposed method can improve the usefulness of the existing method.

抄録全体を表示

As an important research direction in the field of sensors, multi-sensor data fusion has received greater attention and development in areas such as robotics and autonomous driving. This paper provides a comprehensive introduction to the physical model-like and parameter-based data fusion algorithms that are often used in current engineering. Meanwhile, the process, steps and recent developments of the weighted average method and the extended Kalman filter method are highlighted, and multi-sensor data fusion experiments are conducted for each of the two algorithms. The simulation results prove that the data fusion algorithm has a good fusion effect.

抄録全体を表示

As a most important index in calculating network interrelationships, there are many definitions and more than 400 different centrality dimensions, such as degree, and betweenness, among many others, have been developed. All centrality indexes are calculated using the number of connection lines, and its position in a given network. In the automotive industry, interrelationships among partner firms are tied within a typical systemic network known as keiretsu. As it is widely known that different partners play different roles in assembly line, it is crucial to measure the centrality of transaction network in the keiretsu.. Thus, the relative importance of each partner in a connection line in a transaction network should be measured based upon the pivotal nature of each part, which has never been proposed. This paper contributes and advances our knowledge to the literature by proposing a new parts-importance weighted centrality model.

抄録全体を表示

Uncertainty is ubiquitous in data and constitutes a challenge in real-life data analysis applications. To deal with this challenge, we propose a novel method for detecting anomalies in time series data based on the Autoencoder method, which encodes a multivariate time series as images by means of the Gramian Angular Summation Field (GASF). Multivariate time series data is represented as 2D image data to enhance the performance of anomaly detection. The proposed method is validated with four time-series data sets. Experimental results show that our proposed method can improve validity and accuracy on all criteria. Therefore, effective anomaly detection in multivariate time series data can be achieved by combining the methods of Autoencoder and Gramian Angular Summation Field.

抄録全体を表示

Low-power laser therapy is widely used in the treatment of various diseases as a clinical adjuvant therapy. In this paper, a multi-channel low-power laser physiotherapy equipment design scheme is proposed. The laser exciter used in this scheme can generate laser light with a wavelength of 650±20nm. In order to realize the requirement of multi-point irradiation, the device has designed 8 laser channels, and these channels are controlled by the main control unit composed of STM32F407ZGT6 single-chip microcomputer. The user can control the power, frequency and working time of the laser generator only by making some simple settings on the operation unit. In order to ensure the integrity of the treatment process, the system has designed a power-off protection mechanism, which can automatically save the working parameters in progress.

抄録全体を表示

There are individual differences in human cognitive function, and that is a widely known fact. A hypothesis we made is that; giving both visual and audio stimuli may make it easier for people to catch information. In this research, three indicators are set consisting of memory, understanding, and concentration for an experiment. The difference in learning effect due to the reading situation was measured. We concluded that a Text-to-Speech (TTS) and the highlighting system can help reading in some cases.

抄録全体を表示