Automated crane operation at construction sites requires generating a three-dimensional map of the crane's surroundings to determine a safe lifting and carrying route. As a method for generating a 3D map, a disparity image generation method has been developed using moving images acquired from a monocular camera, which is an inexpensive measurement device. However, there is a problem that the disparity of the same point or the disparity of two points differs between the generated disparity images, which indicates that the disparity values have different meanings and cannot be simply overlapped. In this study, we theoretically explain the situations when such cases occur and propose a method to correct and integrate disparity images. The experimental results using simulated and actual moving images demonstrate that the proposed method properly detects reference planes, estimates the parameters required for integrating the corrected disparity images.

 This study investigates learning methods for color transformation models of specific objects under challenging conditions with limited training data, focusing on applications in landscape architecture. Using a small dataset of pedestrian bridge images captured on-site and employing CycleGAN as the color transformation model, we conducted parallel survey research to examine the learning effectiveness of loss function weighting in specific object regions and methods for identifying effective model parameters during the training phase. The results demonstrate that loss function weighting successfully generates model parameters with superior transformation performance. Furthermore, we established that utilizing the Structural Similarity Index Measure (SSIM) as an image similarity metric, combined with the CUmulative SUM (CUSUM) method for change point detection, provides an effective approach for identifying optimal model parameters.

 The utilization of information and communications technology in snow removal to ensure safe and accessible winter road traffic is progressing. However, despite the essential role of snow depth measurement in snow removal patrols, the efficiency of these patrols has not advanced sufficiently. Thus, in this study, we performed experiments using a low-cost, self-developed mobile mapping system (MMS) to identify the snow depth differences in three designated snow-covered sections, i.e., sections A: ≤3 cm, B: 6–8 cm, and C: ≥10 cm. The snow depths were calculated using two methods, i.e., the difference in DSMs and the estimated road surface position. The results demonstrated that both methods successfully identified the differences in the snow depth between the sections. In addition, the results of a statistical analysis demonstrated that the proposed system can detect statistically significant snow depth differences when the difference was 2 cm. These results demonstrate the potential of the low-cost MMS for effective snow depth measurement on roads.

 In this study, we proposed a system that estimates the disaster waste generation caused by active fault earthquakes as soon as possible after a disaster. In this system, seismic intensity information is automatically collected from the Japan Meteorological Agency's website. The system also automatically calclates damage level of each residential building and disaster waste amount of each waste type. Users of system can obtain the detailed generation amount, composition, and spatial distribution of disaster waste in affected cities without complex manual operation. In order to evaluate rapidity and accracy of estimation in proposed system, we applied it for damaged towns by The 2024 Noto Peninsula Earthquake. The results showed that the total amount and spatial distribution of disaster waste could be predicted in each waste type within about 30 minutes of the disaster occurrence. In addition, The result of accuracy evaluation showed the total amount of disaster waste could be calculated reasonably volume.

 I-Construction 2.0, which aims to improve productivity by utilizing 3D data throughout the entire construction and production process of port construction, is being promoted, and the spread of multi-beam sounding, which can obtain detailed topographical information over a wide area, is expected. On the other hand, it is difficult to perform simple sounding using manned vessels, and the data processing method carried out by experienced engineers based on their experience has the problem that the labor required for analysis processing increases as the amount of data increases compared to single-beam sounding. To solve this problem, this study verified the application of multi-beam sounding using an autonomous small vessel and automated data processing in the field. Specifically, it was demonstrated that by combining an autonomous small vessel with automatic processing such as the CUBE algorithm, it was possible to obtain highly accurate water depth values while significantly reducing the labor required for surveying and data analysis, and that construction management was made more efficient.

 To measure the completed form of structures, we verified the depth accuracy of point clouds from terrestrial laser scanners (TLS) and SfM/MVS from videos using flat targets made of materials used in stations. For TLS, material, distance, angle, and environmental conditions affect accuracy, with TLS model, distance, and laser angle being key factors. For SfM/MVS, the shooting device, distance, and angle between the camera and subject affect accuracy, with materials having many good feature points, distance, and angle being key factors. These factors must be considered when selecting and performing point cloud acquisition methods.

 Many utility tunnels were constructed during the period of rapid economic growth. There is a need for appropriate maintenance management to ensure their longevity. The damage information for maintenance is managed through documentation and photographs. However, the interiors of utility tunnels often have similar structures, making it difficult for inspectors to identify the locations of damage. Moreover, there is currently no support system to assist this challenge. This study aims to develop a system for accurately identifying damage locations in utility tunnel inspections and enhancing the communication of damage information. Multiple measurement instruments were used to measure the utility tunnels and construct 3D point cloud data. The utility tunnels were visualized using 3D data and 2D maps. The damage location and type were indicated through the 3D data, while the condition of the damage was shown using omnidirectional images. The proposed system was evaluated by inspection engineers and suggested the capability of the system.

 This paper presents a hybrid MR (Mixed Reality) visualization system that enables seamless and highly accurate superimposition in any outdoor environment by selecting two types of superimposition methods: location-based and marker-based. The system uses HoloLens 2 as the MR device, and performs location-based positioning in an open-sky environment with good GNSS (Global Navigation Satellite System) receiver reception, and automatically switches to marker-based positioning in a non-open-sky environment with poor reception. An opening model is also introduced to visualize underwater structures naturally and quantitatively. To verify its validity and effectiveness, the present MR visualization system is applied to the visualization of the base of piers of a road bridge.

 Current automobile traffic volume surveys manually count the number of vehicles passing through the surveyed cross section, which limits the survey days and times. In recent years, services and methods have been developed to automatically count the number of vehicles by vehicle type from video images captured by video cameras, but they are difficult to apply in low-light environments. However, it is difficult to apply these methods in low-light environments. In this study, we developed a method to count the traffic volume by vehicle type using deep learning from point cloud data measured by non-iterative LiDAR, which is inexpensive and can comprehensively measure the measurement range. As a result, the F value for counting the number of cars was more than 0.930 at four points, and the F value for classifying small cars was more than 0.889. These results indicate that inexpensive LiDAR can be applied to automobile traffic volume surveys.

 In our country, traffic analysis using ETC2.0 probe data is generally conducted on a route-level basis, while lane-level traffic analysis is rarely seen. Existing studies have aimed to enable lane-level traffic analysis by overlaying polylines connecting the positioning points of ETC2.0 probe data onto a mesh and using machine learning to estimate individual vehicle lanes. However, in the curved sections of weaving sections, the polylines could not be overlaid onto the mesh, making it challenging to apply the estimation methods. This study proposes a method for estimating individual vehicle lanes that can be applied even to the curved sections of weaving sections. The validation results demonstrated the potential to estimate vehicle lanes by correcting the polylines in the curved sections. Furthermore, by expanding the width of the mesh, it was shown that it is possible to estimate vehicle lanes even for ETC2.0 probe data positioned outside the lane.

 In recent years, using surveillance cameras installed on streets and commercial facilities, pedestrian safety and customer behavior analysis have been focused on. However, individuals are mostly detected and manually tracked from recorded videos. As labor and costs increase with growing number of cameras, image processing technology that automatically indentify individuals using deep learning have spurred interest. Current technologies mainly focus on reidentifying specific individuals registered in the database system, overlooking unregistered or unknown individuals. Therefore, in this research, we develop the method that uses machine learning to capture the relationship between the features of a person obtained by deep learning, identifies known individuals, and adds unknown individuals to the reference set. This method enables person identification among multiple cameras.

 This study proposes an automated track center estimation process from Terrestrial Laser Scanner (TLS) point cloud data, focusing on the “characteristic arrangement of buildings and structures along railway tracks in railway spaces.” Unlike Mobile Mapping System (MMS), TLS does not provide trajectory information along tracks that could serve as a clue for identifying track areas. To address this challenge, we established solutions through the following approaches. First, we estimated track extension direction using the first principal component of local point clouds. Next, we performed rail detection from point cloud cross-sections using deep learning. Finally, we developed a robust track center estimation process. We verified our method using TLS point cloud data from field observations of railway spaces and confirmed the effectiveness of the railway space arrangement characteristics, which was the focal concept in method development. These results contribute to resolving bottlenecks toward automating the construction of digital models for railway spaces using TLS point cloud data.

 In recent years, the traffic accidents caused by intentional driving operation errors or dangerous driving have become a serious social problem in Japan. While the technology is being developed to automate driving which the driver does not have to operate the vehicle, there are still many issues to be resolved, and it is expected that it will take some time before it is introduced. Thus, it is essential to continue to development the preventive safety technology that supports the driver's operation assitance and prevents accidents.

 In this study, the anomaly detection technology would be effective as one approach to the development of preventive safety driving technology, and it proposed the method of threshold of anomaly driving detection that takes into accout the heterogeneity of individual drivers. Anormaly detection often occurs in false positives, and the incorrect control and warnings that result from these false positives not only cause excessive stress for the driver, but also contribute to a decrease in the reliability of the preventive safety technology itself. Therefore, it is important to set thresholds based on the each individual driver, and in this study, the each proposed setting methods was verified for the two types of anomaly driving detection methods.

 GIS can currently load three-dimensional building models online, but most of the models are LOD 1 block models without textures. In this study, in order to improve the accuracy of the model shape from the block shape of LOD 1 to the opening shape of LOD 3, the dimensions of the building components were measured using TS opposite side measurements and the laser distance meter, and these measurements were used to model the openings, pilotis, pillars, verandas, windows, rooftop, and exterior wall exhaust vents using GIS. In addition, the side view photographs were attached in order to apply high accuracy textures to the model. The aspect ratios for distortion correction were determined from the actual measured values. As the result, we were able to create the model with highly accuracy textures at LOD 3.

 Markerless AR can be used indoors and outdoors without the need for markers, so it is used in a wide range of fields such as navigation, product sales promotion, building maintenance and management, and medical care. There is a particularly high need for AR representation of GIS data such as drawings. The authors have developed a markerless AR system that uses SLAM to automatically generate markers from planar positional relationships (distance and angle), but there was a problem with this system in that the position of the superimposed representation would shift when the device was moved. In this research, we propose a technology that sequentially corrects the problem of the superimposed representation during movement using one of the new planes detected by SLAM. Experimental results showed that the proposed method can sequentially correct the problem of superimposed representation.

 Dynamic maps are based on high-precision 3D maps created using 3D point cloud data acquired by MMS equipped with GNSS. In the future, it is planned to develop dynamic maps for general roads where vehicles turn frequently, such as intersections, and there is concern that the reliability of the point cloud data will deteriorate. In this study, we compared the point cloud accuracy between straight sections and turning sections, and quantitatively verified the effect of the combination of GPS, QZSS, GLONASS, and Galileo among GNSS on the accuracy of 3D point clouds acquired by MMS. The results showed that the point cloud accuracy was lower in the turning area than in the straight area. The horizontal accuracy of the point cloud was not significantly affected by the combination of satellites, while the height accuracy was more stable when Galileo was used.

 In recent years, the increasing need for countermeasures against escalating disasters and the momentum for digital transformation in the civil engineering industry have led to the development of a platform. This platform aims to contribute to disaster prevention measures for road networks through the interconnection of numerous publicly available data sets, with xROAD's bridge information as the central component and other public data integrated via API. The system's dashboard and filtering functions enable macroscopic analysis of a large group of bridges, while data overlay features such as registered 3D models allow for microscopic analysis of individual bridges. When this system was applied to field surveys and damage analysis of bridge damage during the 2024 Noto Peninsula Earthquake, it demonstrated the effectiveness of rapid information collection on bridges during field surveys and the utility of bridge damage analysis through the overlay of multiple data sets. Additionally, the verification of road risk assessments suggested its potential future application in risk assessment management.

 In order to prevent damage to previously buried objects during underground infrastructure construction, there is a need for technology to detect the location of underground objects using ground penetrating radar. Ground penetrating radar irradiates electromagnetic waves into the ground from the surface, uses the reflection from buried pipes, and measures the earth cover of buried pipes by interpreting the image. In this study, measurements were taken at a test yard where the locations of buried pipes and cavities are known, using the VRS GNSS receiver that can perform positioning with an error of several centimeters and the ground penetrating radar equipped with a distance measuring instrument. The position detection accuracy was determined by examining the shape, size, and depth of the cavity, as well as the possibility of detection. As a result, the accuracy of position detection of buried pipes and cavities linked to VRS positioning and distance measurement was within 0.2 m for horizontal range and approximately 0.05 to 0.2 m for vertical range. In addition, the vertical range increased as the depth increased.

 Among the real-time spatial understanding methods using SLAM (Simultaneous Localization and Mapping) technology, self-positioning using Lidar SLAM technology uses lidar, and uses a pre-measured base map without using GNSS or IMU. This is done by matching the lidar point cloud with the lidar point cloud at the time of measurement. However, the factors that affect the accuracy of self-location estimation have not been sufficiently studied. In particular, there are few cases in which the effects of the traveling speed of the measurement vehicle and the point cloud search range have been investigated. In this study, we conducted an experiment focusing on the influence of the moving speed of the measurement vehicle and the point cloud matching range in self-position estimation using Lidar SLAM. By comparing the results with positioning results from high-precision GNSS/IMU, we quantitatively examined the effects. As a result, it was found that there was almost no effect even if the traveling speed was different from 20 km/h, 40 km/h, or 60 km/h, and that the accuracy of the point cloud matching range decreased significantly when neighboring points were included.

 The Centimeter Level Positioning and Augmentation Service (CLAS) provided by QZSS has not yet established an effective solution checking method. In this study, we conducted an inspection using CLAS in accordance with the inspection method stipulated in the work regulations for the network-based RTK method, and examined its effectiveness. The results were compared with the Boxplot diagram and the inspection correctness rate. As a result, the inspection correctness rate was more than 75% for the X coordinate and more than 90% for the Y coordinate, but did not reach 70% for the elevation, when the allowable range was the same as that of the standard operating rule. Therefore, the inspection method verified in this study may be effective in the horizontal direction (X and Y coordinates), but it is difficult to apply it to the inspection of elevation.

 In recent years, the measurement of 3D point clouds using the Mobile Mapping System (MMS) has been attracting attention, and MMS can efficiently acquire 3D point cloud data. In shielded environments where GNSS cannot be received, such as tunnels, the position information is supplemented by an Inertial Measurement Unit (IMU) to estimate the position and orientation of the MMS. Existing studies have shown that the longer the measurement distance in non-satellite receiver space, the larger the range difference due to the IMU. Therefore, in this study, we conducted mobile measurement using MMS in a tunnel where no satellite reception is available, changing the driving conditions, and evaluated the position error at the verification point. The results of the test under a variety of running conditions (10 patterns, including straight ahead, center turnaround, and end turnaround) confirmed that the measurement performance was within 0.1 m in 3D range for up to 60 seconds after tunnel entry. Since the period of no satellite signal reception in this experiment was very short (less than 60 seconds), future verification will be conducted in environments where satellite signals cannot be received for long periods of time.