The authors have been developing a method to assess the building damage at the district level in heavily damaged areas using satellite-derived SAR data, and conducted the coherence analysis of urban areas. In this study, we focus on the Kumamoto earthquake to clarify the relationship between the coherence and the building damage rate using ALOS-2 PALSAR-2 images in the L-band under multiple acquisition conditions (different orbits, looking directions). The analysis area was Mashiki-Town, Kumamoto Prefecture, and we evaluated the relationship between the average coherence value and the building damage rate within a 200-meter mesh in Mashiki-Town. The results showed that there was a significant negative correlation regardless of the acquisition conditions, and revealed that the smaller the building damage rate, the larger the coherence variation. Three indices of the building damage were examined based on the field survey data, but no significant differences were found among them.

 Recently, labor shortages at construction sites have become more serious, and improving labor productivity has become an issue. One of the solutions to this problem is the automatic operation of cranes, and it is considered that a method of 3D reconstruction around the crane using video images obtained from a monocular camera attached to the top of the crane boom is effective. However, the crane hook is always reflected in the images obtained from the camera, and this is a factor that significantly impairs the accuracy of the 3D reconstruction. Therefore, in this study, we attempted to discriminate objects in the image in order to remove such objects from the image, which may cause a loss of accuracy in 3D reconstruction. Considering that this is a preliminary step of the 3D reconstruction process, we attempted to classify objects based on feature points rather than whole pixels in order to reduce the computational cost as much as possible. We proposed three classification methods based on the fact that each object has its own characteristic optical flow trajectory. As a result, the method that incorporates Bayesian statistics in the classification is robust to the effects of camera vibration, and the method that incorporates Bayesian statistics in the classification is robust to the effects of camera vibration, The proposed classification methods are effective except for the case when the crane is stopped and the trajectory of the optical flow is obtained as a fixed point, in which case Accuracy, a measure of classification accuracy, is more than 0.85.

 In recent years, Japan has witnessed frequent occurrences of extreme winds and floods, as well as a consequent increase in the number of cases of people in vehicles being stranded in a flood, particularly in underpasses, which are generally found in urban areas. Underpasses are prone to flooding because their elevation is lower than that of the surrounding areas. In this study, we analyzed evacuation routes under normal conditions and flooded underpasses using data from road networks and evacuation centers. We quantitatively evaluated the difference in evacuation distance during flooding and its causes. The results of the analysis on the Osaka Prefecture showed that urban areas were less affected by the flooding because of their high density of evacuation centers; however, in few regions, evacuation routes needed to be detoured more than eight times compared to that of normal conditions. In some areas, the nearest evacuation center differed between normal and flooded areas. In case of flooded underpasses, it is important to consider multiple evacuation routes as a precaution.

 As labor shortages at construction sites have become increasingly serious, technologies such as crane operation assistance and fully automatic operation have been attracting attention, and accurate and fast 3D mapping of the crane's surroundings is indispensable to realize such technologies. A camera installed at the end of a rotating crane boom has a disadvantage in terms of photogrammetry compared to a camera moving in a straight line. Therefore, this study attempted to generate a 3D map around crane by generating a disparity image using stereo matching technology. The accuracy was improved by projection transformation of the disparity image from a central projection to an orthographic projection, automatically setting the disparity threshold, and overlapping multiple disparity images. As a result, a dense 3D point cloud, which was difficult to obtain in previous studies using SLAM, was obtained in quasi-real-time processing time.

 Recently, convolutional neural network (CNN) based model for determining the rust condition rating model for weathering steel bridges has been proposed, and many studies have been carried out for practical application. The goal of this study was to improve the versatility of the rust condition rating model by diversifying the images of rust, which is the training data. The accuracy of the rust condition rating model by using CNN when trained on angled rust images was verified, and the versatility of the proposed model for rust images taken by inspectors under various conditions was investigated. The results showed that the accuracy of the proposed rust condition rating model increased by training rust images taken from various angles.

 Break lines detection using 3D point cloud data is used in various fields such as shape recognition of structures, CAD data creation, downsampling, and plane (TIN) generation. In the previous research to detect break lines, a method using plane intersection lines of point cloud data, a method using normal vectors, and a method to find and connect cross-sectional change points for each cross-sectional view have been proposed. However, there are problems such as the inability to detect accurate break lines due to the noise of vegetation and trees included in the 3D point cloud data, and the limitation of target structures. Therefore, in this research, we propose a method for detecting break lines of structures that is superior to previous research by using the characteristics of the RANSAC.

 In recent years, such as the 2018 Northern Osaka Earthquake, there have been some cases in which railway operations have been stopped for a long time for inspection, although there is almost no damage to the railway structure. Here, by utilizing technologies such as sensing, which have developed remarkably in recent years, it is conceivable to save labor in inspections and achieve early recovery. However, there are a wide variety of inspection points on railways, and it is not easy from a cost standpoint to attach sensors to all the target points. Therefore, in this paper, we propose an efficient sensor placement method for railroad tracks, taking into account inspection time and travel time, targeting bearings as sensor installation locations. By adopting the proposed method, it was verified that the layout for effectively reducing the inspection time can be evaluated while greatly reducing the calculation time compared to the case of performing a round-robin calculation.

 In this study, the authors aim to show an analysis method that can grasp the actual state of pedestrian behavior of tourist and congestion in the street space of a tourist destination using Wi-Fi packet sensor data and to clarify the impact of street usage on pedestrian behavior for different purposes. Through the analysis of the Higashiyama area in Kyoto, the authors demonstorate that: (1) Visualization of the number of hourly observation data by seasons was conducted to understand changes in the use of the link in response to measures for dispersion of tourists, such as lighting-up of famous signtseeing spots. (2) A mixed normal distribution assuming multiple purposes of tourists was estimated using a constrained EM algorithm, and the characteristic values of the distribution corresponding to simple movement, short-time stay, and longtime stay were determined. Furthermore, the time variation of the values was used to understand the tourist behavior and its situation in each link. (3) Based on the relationship between the number of observations and the mean value of each distribution for each purpose, the tourist behavior and environmental factors that affect travel time, as well as the level of congestion that suppresses purchasing behavior, were clarified.

 Since social infrastructure, which was intensively developed during the high economic growth period, will deteriorate all at once in the future, maintenance and management of facilities will be an issue in the future. Currently, facility inspection records are based on paperbased forms, and are not premised on automatic processing by computer. The authors have developed the “Smart Chosa” and realized a database of facility inspections and a GIS System. The Smart Chosa was able to record the location of the inspection photo on a two dimensional map, but because it was necessary to approach the deformed part when taking the inspection photo, it was not possible to grasp the position, direction, and size of the entire facility. There was a problem. Therefore, in this paper, we apply 3D models to the Smart Chosa for sabo dams. The above problem was solved by inputting the 3D model of the sabo dams generated from the photograph taken by UAV into the 3D GIS and pasting the photograph taken on site to the 3D model of the sabo dam.

 The frequency of torrential rainfall has increased in recent years. Japanese rivers have steep slopes, short lengths, and small basin areas. Therefore, the risk of flooding is high, and understanding the river channel shape and cover as a basis for disaster prevention measures is essential. In terms of the environment, environmental maps of first-class rivers in Japan are prepared based on the census of riverside areas and are required to be updated once every five years. We believe that changes in biomass, which quantifies the amount of greenery, can be used as basic data for implementing various measures for both disaster prevention and the environment (e.g., changes in the runoff function focusing on roughness, changes in the amount of greenery in terms of CO₂ absorption, etc.). In this study, we aimed at estimating the biomass of a first-class river in Japan and its long-term evolution (about 40 years) and devised a simple method to extract grasslands and forests from optical satellite images. Based on the results of this study, we proposed an excellent method for practical use by devising a method to extract green areas (herbaceous and tree) focusing on the quantization level of the data and the season.

 Global warming is considered the main cause of rising sea levels. Flat atoll islands in the South Pacific, with an average elevation of only a few meters, are affected by rising sea levels, especially in the coastal zone. Hardware measures, such as installing levees and shore reclamation, are important for adapting to climate change in these islands. Additionally, the protection of foraminifera contributes to island formation. However, foraminifera is decreasing due to population density. Therefore, it is crucial to study the foraminifera habitats and take measures, such as establishing protected areas. In this study, new aerial photography and surveying were conducted based on preliminary survey results. Foraminifera habitat areas were then estimated using image analysis combining aerial photographs and geospatial information. Correspondence with foraminifera habitat potential and classification accuracy tables reveals that foraminifera habitat areas could be estimated.

 One of the challenges in underwater acoustic positioning is the existence of large errors in multipath environments. We have developed a new underwater acoustic positioning technique with signal filtering technology to eliminate unnecessary reflected waves. This technique is based on a computer calculation of the propagation time from a sound source to each receiver, and selects only the time-of-flight of the direct wave to stably and accurately measure the baseline length even in multipath environments. In this paper, we evaluated the performance of the proposed method in a mobile environment. In a large pool with multipath conditions, a boat with a sound source was moved and its position was measured in real-time. The average positioning error of the moving object was 0.12 to 0.24 (m), and the missing rate was 0 %.

 In this paper, we focused on a location-based AR that does not require marker installation, and developed a visualization system based on location-based AR using small GNSS receivers. We compared the acquisition accuracy of position information in an open-sky environment and a non-open-sky environment. We developed a simple and highly accurate superimposition method by correcting the angle using two receivers. In order to examine the validity and effectiveness of the visualization system, we applied the system to the visualization of urban river flow and compared the result obtained by the conventional marker-type AR visualization

 Illegal dumping in the river is grasped by the river patrol once or twice a week, and the minimum size is about 30 cm. Since it is difficult to grasp illegally dumped wastes in low waterways that are difficult to see from embankments, it is necessary to comprehensively grasp illegally dumped wastes in the entire river area including low waterways. Compared to satellite and aerial photographs, UAV photographs are considered to be more suitable for river patrol because they are easier to take and have higher resolution. However, it takes time and effort to visually interpret illegally dumped waste from UAV photographs. Therefore, in this study, in order to realize a comprehensive grasp of the entire river area, we confirmed the feasibility and effectiveness of work efficiency improvement of the illegal dumping grasping method that combines UAV photography and AI. As a result, we were able to grasp the comprehensive distribution of illegal dumping in the entire river area from the UAV photograph, and it was confirmed that the automatic detection by AI shortened the work time to detect illegal dumping.

 In Japan, the percentage of sewer pipes that have reached the standard service life of 50 years is rapidly increasing. Therefore, there is a need for an efficient maintenance and management methods for sewer pipes. However, the current inspection process is time-consuming and labor-intensive because the surveyor visually checks for abnormalities in the images of the sewer pipe taken by CCTV camera. Furthermore, the surveyor determines the degree of damage based on experience, making uniform and quantitative evaluation difficult. In this research, we propose a method to detect damage such as cracks and breaks and to determine the degree of damage by using deep learning to analyze video images taken of sewer pipes. This enables quantitative evaluation of the degree of damage and contributes to reduction of labor requirements and to the advancement of inspection work. We confirmed the usefulness of the proposed method through empirical analysis.

 In recent years, the number of heat stroke patients and deaths has increased due to rising temperatures caused by global warming and the heat island effect. Heat indices established to determine the likelihood of heat stroke are used to publicize the risk of heat stroke, and the Ministry of the Environment publishes heat indices for each city on the Web. However, there is no way for the general public to know the local heat index in outdoor environments such as daily life and workplaces because specific measurements are required to measure the radiant heat necessary for the heat index. Knowing the risk difference between sunny and shade distribution at any given time will facilitate planning actions and tasks. This paper quantitatively demonstrates the real-time visualization of heat risk for public parks and work site landscapes. In this study, we calculated the sunlight conditions based on 3D data from the site in order further to estimate the heat index on a pixel-by-pixel basis using a global illumination shading representation. By implementing this method in a game engine, it was possible to develop a system that visualizes the heat index distribution in real time for ever-changing heat environments.

 Under the COVID-19 infection, consumer visits to tourist attractions and commercial establishments declined, which took a heavy toll on the economy. On the other hand, avoiding crowding prevents transmission of not only COVID-19 but also other infectious diseases, and the density of people in public and commercial facilities is likely to continue to affect the behavior of citizens. It is essential to reduce the risk of congestion without restricting people's behavior, and there is a growing demand for information on congestion levels. Existing technologies that visualize congestion by color-coding using motion sensors have the disadvantage that the visualized content could be more abstract, making it difficult to grasp the congestion situation. This study proposes a method to visualize the distribution of people while moving around the site using images captured by a 360° view camera. SfM can reconstruct the 3D shape of the target space and the shooting viewpoint, and a machine learning discriminator is used to extract and track people and map them into the space. This paper demonstrates the visualization of human crowding levels at various locations on a university campus.

 In recent years, point cloud data measured by mobile mapping systems (MMSs) haves been used for the maintenance and management of roadside structures. However, using MMSs to perform frequent routine inspections is not cost-effective. To this end, we previously developed a sensing unit mounted on a car using inexpensive sensors. A self-positioning correction method using simultaneous localizatioiin and mapping with multiple lidar detection and ranging (LiDAR) sensors installed horizontally and diagonally was investigated, and its usefulness was confirmed. However, the diagonal installation of LiDAR has a problem in that the accuracy of point cloud data generation deteriorates when the measurement data between consecutive points have the same shape. Additionally, it is necessary to superimpose multiple LiDAR point cloud data with different measurement ranges for a comprehensive measurement. Therefore, here, we propose a method to generate point cloud data over a wide area by using the self-position of the horizontally installed LiDAR to correct the self-position of the diagonally installed LiDAR and superimpose the point cloud data of both installation methods. In the result, the method was verified in some experiments, we confirmed it was useful.

 In Japan, block walls are commonly used to separate structures.They exhibit excellent safety and security features, but may collapse in the event of a disaster, impeding the passage of emergency vehicles and the evacuation of people. To address this problem, we propose a technique to extract the position and inclination of block walls automatically based on point cloud data, with the aim of facilitating their maintenance and management.Further more refinement of this technology to evaluate the possibility of blockage of adjacent roads caused by the collapse of a block wall based on point cloud data is expected to enable the development of more appropriate evacuation routes. Therefore, in this study, the authors develop a technique to determine road closures caused by collapsed block walls based on point cloud data.

 In mountain tunneling, many fatalities occur in the tunnel face after blasting due to falling rocks, and the face should be unmanned as soon as possible. Automated machines and remote-controlled machines have already been introduced to construction sites to reduce manpower and improve safety, but there are no examples of unmanned or automated face scaling. On the other hand, unmanned construction, which began in 1993 in response to pyroclastic flows and mudslides caused by the eruption of Mount Unzen-Fugen, is now being implemented by local construction companies in some cases, and is expected to become widespread. However, automated operation is still difficult. In this study, aiming at the spread of automatic operation in construction sites, we developed a method to automate existing construction machinery and applied it to automatic driving and automatic scaling in mountain tunnels, which are GNSS-Denied environments, by utilizing SLAM (Simultaneous localization and mapping). We also verified its practicality.

 In Japan, social infrastructure constructed during the period of rapid economic growth is seriously aging, and there is therefore an urgent need to inspect and repair it to extend its service life. Accordingly, the Ministry of Land, Infrastructure, Transport, and Tourism is promoting CIM and i-Construction with the aim of improving the efficiency of maintenance and management operations by utilizing point cloud data and 3D models. However, it is difficult to measure the entire structure of an elevated bridge or a road with heavy traffic because of the limited measurement locations. In this regard, existing research has been unable to generate 3D models from such incompletely measured point cloud data. Therefore, we propose a novel approach that uses deep learning to generate a parametric bridge model to complement missing point cloud data.

 In public works projects in recent years, a lot of point cloud data have been measured and accumulated in various places by introducing mobile mapping systems, ground-mounted laser scanners, etc. Since point cloud data are the aggregation of large amounts of points, use is limited if they have not been processed. Therefore, the authors gave the meaning of road features such as traffic lights and signs to point cloud data extracted using dynamic maps to propose a method to construct product models capable of spatial processing and retrieval in road feature units. Meanwhile, the scope of application of the proposed method is limited in zones where the completed plan view and road map have not been maintained. Using the point cloud data of road features automatically extracted from the proposed method, this research created training data for automatically identifying road features to construct the identification models of road features. As a result, the road features could be automatically identified from the point cloud data in zones where the completed plan view and road map have not been maintained.

 In our country, the utilization of ICT is promoted toward streamlining and labor saving in the investigation of the traffic volume of automobiles. In particular, investigation methods using moving images are drawing attention and various methods have been proposed. Now, the counting of a spot traffic volume has been realized but the flow investigation of vehicles between points has not been realized. Therefore, by adding a technology to identify the same vehicle to the existing technology, the investigation of the flow of vehicles between points becomes possible and this contributes to the sophistication of the investigation. Accordingly, this research created a method to identify the same vehicle by recognizing a series of designated numbers on license plates from moving images for a traffic volume investigation shot with more than one wearable camera. In the experiment, the created method was applied to moving images shot between two points to verify whether the same vehicle could be identified. As a result, the possibility was clarified that the same vehicle could be identified, which could not be measured with existing technology.

 For safe, secure, and smooth road space maintenance, it is necessary to understand traffic reality changing from moment to moment. Therefore, road administrators analyze road traffic using probe data acquired from vehicles. As a concrete example, after map matching processing for identifying a road where probe data have been acquired using road network data in which the road is represented as a line segment, a travel speed in each road unit, etc., are analyzed exhaustively. Meanwhile, in the analysis of a speed for each movement direction for each intersection, etc., map matching and counting often cannot be processed mechanically. This research created a method to determine the movement directions of probe data at an intersection using polygon meshes for understanding a positional relationship only with the latitude and longitude. Accordingly, a demonstration experiment was conducted, the percentage of correct answers for movement directions was 90% or more and knowledge was obtained that movement directions can be determined mechanically using the created method.

 It is important to count the number of people in a crowd during an event and commuting hours to prevent accidents. In recent years, a method has been developed to count the number of people easily from images by improving the speed and accuracy of deep learning. However, since crowd shooting conditions such as the installation angle and height of a camera are varied, depending on the size of a person and the degree of occlusion in a moving image, also how the person seems is varied. Thus, it is difficult to accurately count the number of people in a crowd under various conditions using one counting method. Consequently, it was considered that a counting method could be established to secure a certain degree of accuracy by categorizing scenes and conditions for shooting a crowd and changing a method to count the number of people best for the status of the crowd in each scene and shooting condition as necessary. In this research, four types of methods to count the number of people were applied to each scene to clarify the best method for each scene and problems toward the changing of an applied method.

 Japan is covered with mountains and hills, and it is a problem that there are many disasters such as landslides caused by earthquakes and torrential rains. At present, the judgment of the sediment disaster section is done by visual interpretation work from the aerial photograph taken after the disaster. However, this work requires a long period of visual observation by an experienced person, which places a heavy burden on the worker. In order to improve the efficiency of this work, the study on automatic detection of sediment moving parts by deep learning using aerial photographs has been advanced in recent years. In this study, in order to improve the accuracy of automatic detection of sediment moving parts by UNet, the authors applied UNet++ as an image segmentation model, which has the feature of suppressing false detection and is expected to improve the accuracy of sediment moving part detection.

 This paper analyzes the impact of intention to promote bus use and regional exchange by information provision. Especially this research is focusing on to browse pictures in the bus car which was taken in several decades ago in the region of bus route. This paper describes the analysis results using interview survey data which was conducted in Yusuhara town, Kochi prefecture by showing the pictures of Yusuhara town to respondents. Based on the collected data, structure of intention on bus use and regional exchange was analyzed by applying structural equation model. Structural equation model is developed by following the structure of AISAS model which is describing the flow of intention from Attention, Interest, Search, Action and Share. Estimated model showed the possibility of bus use promotion by introducing picture browse service. Regarding on intention of regional exchange by the picture provision, estimated model described that intention to tell for sounding people such as family, relatives and neighbors possible to be also promoted. This paper showed the possibility of picture browsing services in the bus car to promote on both bus use and regional exchange which is important factors to revitalized public transportation use at mountain and rural area in Japan.

 In recent years, natural language processing technology has made great strides with the advent of BERT, but it has been pointed out that there are issues in applying BERT to specialized fields where the amount of data is small.The Ministry of Land, Infrastructure, Transport and Tourism Data Platform has developped data aggregation platform that enables various types of data to be collected and utilized to promote various types of innovation and improve efficiency.However, the accuracy of existing natural language processing technology may not be sufficient for the specialized texts that we use in our daily lives, such as construction information, patrol inspections, and technical information in the civil engineering domain.We constructed “civil engineering BERT” by training BERT on sentences related to civil engineering.We verified the accuracy of the constructed “civil engineering BERT” and the existing BERT, and showed the superiority of "civil engineering BERT", and confirmed that training civil engineering sentences is effective.

 Many of infrastructures were intensively developed during period of rapid economic growth, and the increasing demand for maintenance management due to aging is an urgent issue, and more efficient and strategic maintenance management is required. In road maintenance management, two-dimensional data are used. However, it is difficult to grasp the detailed structure of bridges and slopes and inspection points. There is a problem that inexperienced engineers may not have a correct understanding of the on-site situation. In this study, we proposed a information system that measures and visualizes road damage as three-dimensional point cloud data and visualizes it on a two-dimensional map, in order to improve the efficiency of maintenance work and to accumulate data. The system was evaluated by practitioners, and its usefulness and issues were indicated.

 Didital twin projects are underway in urban spaces to link real and cyber spaces. However, there are few efforts to accumulate data for digital twin in the construction field. By defining acquired data such as site progress and environmental information in the construction field, secondary uses such as safety management, quality control, and investigation of the causes of construction defects become possible. In this study, we consider and define data that can be obtained at the construction site as data used for three-dimensionalization of the construction site, and consider a digital twin based on three-dimensional point cloud data that can check the work progress during and after construction. In addition, the issues of data acquisition using multiple devices at construction site and visualization of the construction site on a digital space by superimposing each data will be clarified.

 This paper presents a simulation-based tsunami simulation experience system using VR technology for disaster prevention education. In order to create a regional model with high accuracy and ease, we used GIS/CAD and drone data, and developed a modeling method to create a building model based on rules. We also developed a Unity-based rendering method for high-quality tsunami CG. We integrated the city/regional model for areas where tsunami damage is assumed, the results of tsunami and evacuation simulations, and developed a system that enables experiences from the perspective of evacuation sites and evacuees. In order to evaluate the usability and effectiveness of the system, In order to investigate the efficiency of the method, we applied this tsunami simulation experience system to disaster prevention education for local junior high school students.

 The Centimeter level Augmentation Service (CLAS), a service provided by the Quasi-Zenith Satellite System (QZSS), calculates high-precision corrections using data from GPS-based Control Station maintained nationwide by the Geospatial Information Authority of Japan (GSI), Ministry of Land, Infrastructure, Transport and Tourism, to accurately determine current position. This service transmits centimeter level augmentation information from QZSS satellites to accurately determine the current position. The results were evaluated based on the Fix Rate, RMS error, and other parameters. The results showed that the use of Galileo for both real-time and post-processing analysis tended to reduce the horizontal error.

 In this study, to evaluate the performance of Centimeter Level Augmentation Service (CLAS) on an urban street, we conducted a positioning experiment using CLAS (AQLOC-Light). As evaluation indices, we focused mainly on the Fix rate, the number of satellites, and the consistency between the position of the traffic lane and the positioning coordinates. As a result of the evaluation, it was confirmed that CLAS tends to have large variations in the Fix rate and the Float rate. It was confirmed that the fix solution has a certain consistency with the traffic lane position, but problems such as the occurrence of Fix solutions with low accuracy were also clarified. Fix solutions with low accuracy mainly occurred when the experimental vehicle stopped or when getting Float Solutions with large positioning error, and the positioning position deviated from the traffic lane position. The number of satellites (CLAS) tended to be higher when the Fix solution was got than when the Float solution was got.

 Mobile mapping systems (MMS) are being used to inspect road infrastructure. One of the challenges is how to analyze the data obtained from the measurements. In this study, an MMS equipped with a phase-shift laser scanner was used to measure tunnels and automatically detect tunnel damage using an image processing contour extraction method. As a result, a 4-cm area that resembled a collapse was automatically detected. At the same time, it was found that undamaged areas were also detected incorrectly due to the influence of noise and other factors.

 This study demonstrates that our machine learning-based detection technique performs good classification of multiple types of normal and anomaly values that usually occur due to sensor problems in time series data. Water level data in our agricultural field measurement usually includes typical-pattern (i.e., normal), flood-based, spike noise, and slide-shifted data. We introduced a self-organizing map (SOM) to classify these four-type items at the same time. Then, the four classifications were visualized on the 2D map using three types of clustering methods (K-means, Ward, and Vote). The accuracy evaluation for the classification was performed by Accuracy and f1 score based on the rule of binary classification. The accuracy evaluation shows that the Vote method had better scores compared with the other methods. Additionally, for the Vote method, true values for the four classifications were mostly plotted on the same classification region, determined by clustering methods.

 In recent years, along with the promotion of Society 5.0, advanced technologies such as AI and IoT have been introduced in various fields. Contributions to the streamlining of operations and improvement of safety have been expected at construction sites and interest in the development of technologies to manage the positions and status of workers has increased. The authors proposed a method to identify people by focusing on helmets worn by workers and analyzing patterns attached to helmets through deep learning. However, in previous research, thresholds were set for RGB values to extract the pixels of patterns by image processing; therefore, if the colors of patterns change due to sunshine conditions and weather effects, pattern extraction will fail. Thus, this research created a method to extract patterns for general purposes even if scanning is done under various environments using deep learning toward the improvement of the method to identify people. Then, through a demonstration experiment, knowledge was obtained that the proposed method is useful.

 The Ministry of Land, Infrastructure, Transport and Tourism is promoting Project PLATEAU, measures for the maintenance, utilization and adoption of open data of 3D urban models. In fiscal 2022, 3D urban models for 56 cities across the country were maintained and if the maintenance progresses to indoor spaces, further use expansion can be expected. To realize this, it is necessary to establish a method to create 3D models to which barrier information, etc., have been given by measuring indoor spaces easily.

 This research created a method to create indoor 3D models by the voxel representation of point cloud data. As a result, through a demonstration experiment, the created method clarified that the problem could be solved that the density of point cloud data acquired with LiDAR of a portable terminal was non-uniform and the background was permeable, barrier information such as steps could be determined and data processing was easy as a base for providing various features.

 Microtopographic and topographic maps produced from aerial laser survey data are increasingly used for commercial purposes. Unlike slope gradient maps, wavelet analysis maps can extract terrain steps without being affected by slope inclinations. First of all, we created a simulated terrain with varying slopes and specific height differences to examine those height differences that can be represented by wavelet analysis and slope gradient maps. Then we examined the height differences that can be extracted from the wavelet analysis and slope gradient maps for an average slope of approximately 40°, including the rockfall source, using the results of the field survey. The height difference that could be extracted from the wavelet analysis maps was approximately 2 m, whereas the difference that could be extracted from the slope gradient maps was 4 m or more. Therefore, wavelet analysis maps are better in extracting small height differences compared to slope gradient maps.

 The Ministry of Land, Infrastructure, Transport and Tourism is promoting i-Construction, which utilizes 3D information to improve productivity at construction sites. Most of the 3D information representing the current topography measured by laser scanners is stored as point clouds. Point clouds generally store only 3D coordinate values and reflection intensity, but they can be given RGB values using camera images, and are widely used due to the high visibility of features. However, since the laser scanner and the camera are not integrated and not synchronized in many measuring instruments, moving objects such as moving vehicles appear in the camera image at different times from the time measured by the laser scanner. As a result, there is a problem that a colored point cloud is generated that includes areas that are not colored correctly. Therefore, in this research, we develop a technique to detect and correct defective coloring points that are not correctly colored in the colored point cloud.