For better concrete surface inspection based on image analysis and interpretation, this paper has proposed a Target and non-target image Feature area Classification algorithm (termed “TFC algorithm”). Due to the difficulties of selecting training data sets of various texture features (e.g., fine crack, grain, roughness, stain, etc.) using ultra-high-resolution data in supervised classification, non-hierarchical unsupervised classification procedure is introduced to produce TFC and n-TFC images. The experimental results are summarized as follows: i) By using TFC image together with n-TFC image, candidate areas for inspection can be classified and ranked according to the strength of electromagnetic wave reflection, ii) especially, the surface floating area along borders of fine cracks, that is multiple scattering area, can be extracted and; iii) the prototype TFC system with semi-real-time processing function as well as superior operability has also been developed, which is useful as an idea creation support- and existing system combination-type system.

 Initial Galileo service was launched in 2016, and the convenience of using Galileo is increasing in Japan, so it is expected that Galileo will be used in combination with GPS and QZSS in the future. This study was conducted to inspect the effects of positioning performance kinematic positioning in case GPS/QZSS and Galileo are used together by setting signal strength as a threshold value and selecting satellites to be used for positioning as a study of how to use Galileo. With signal strength mask setting, it was concern that number of sattellites would be insufficient with GPS and QZS alone, but with Galileo was used together, signal strength mask setting had a sufficient effect on the Fix rate.

 Findings from a bridge inspection may contain important maintenance and management information such as the degree of damage, which cannot be identified in the standard categories in the system. In this study, the Structural Topic Model is applied to the findings description, which is a part of the bridge inspection report. The results show that the descriptions contain the type of damage found, and the bridge inspection results can be categorized based on that. Additionally, ASR, a damage characteristic which is one of the early bridge deterioration causes in the Hokuriku region, surfaced as a topic. Consequently, we show that focusing on the ratio of topics in each inspection report can lead to preventive maintenance. By analyzing the findings by professional, it is shown that the system could be enhanced in a manner that does not require increasing the number of standard categories.

 We developed a simulation method of tunnel inspection support system using image analysis called SimCITI: Sim-ulation of Crack in wall Images for Tunnel Inspection. It is being promoted to use the inspection support system in the field of the Infrastructure inspection for efficiency and labor saving. The tunnel inspection systems using image analysis need to handle quantitatively not only the optical properties of their systems but also the optical properties or the image statistics of the objects of the images, tunnel walls and the cracks on the wall in the images. However, the lack of quantitative handling method of optical properties of the walls and cracks has prevented us from developing the tunnel inspection support system using image analysis quantitatively and forces us to check and redesign the system over and over. To make matter worse, it is not certain that how many tunnels do we need to examine the system. In this study, we developed a novel method to handle quantitatively the optical properties of the tunnel walls and the cracks which include the universal properties of the natural images such as the amplitude spectrum of natural images falls inversely with spatial frequency. This leads to an innovative simulation method of tunnel inspection support system using image analysis. We also verify the usefulness of the simulation method by testing it in the simulation of crack detection system using convolutional neural network and the simulation of crack width measuring system using image analysis. The quantitative handling method of the tunnel walls and the cracks turned out to be very useful for both simulation systems. This method, SimCITI, will be highly useful in developing and benchmarking tunnel inspection support systems using image analysis because it makes us prevent from going out from the office to millions of tunnels in the world.

 The bridge images include various information such as the class of the bridge members in addition to the specific damage. The Deep Learning model constructed in this study can generate a description of the damage situation with high accuracy from the bridge images. In addition, the regions that the Deep Learning model focuses on were visualized. It was shown that the Deep Learning model paid attention to the location of the damage and each member when generating sentences. In addition, it was suggested that the Deep Learning model paid attention to not only the damaged members but also the surrounding members.

 The authors have developed an automatic inspection robot for the cables of cable-stayed bridges. The robot moves along bridge cables and captures the cable surface images with cameras. Then, it can generate a panoramic image in order to detect defects on cable surface. However, it is not able to create an accurate panoramic image due to the circumference of the cable. Therefore, the geometric transformation method of a cable surface image is proposed in this study. The method expands the circumference of the cable. The experimental result shows the possibility of circumferential expansion by the method.

 In order to eliminate labor accidents at the construction sites, it is important to implement effective safety management measures. For example, there is a measure that warn a person who enters a dangerous place or contact with construction machines using cameras. Many existing studies have applied deep learning in recent years, which reports that face authentica-tion, gait identification, and human identification lead to results with a higher precision than ever before. On construction sites, however, it may not be possible to apply the existing technology as operators’ clothes tend to be similar to each other and it is necessary to identify construction machines. Therefore, in this research, we focused on the helmet that workers always wear at construction sites, and proposed a method for identifying a person wearing a helmet with pattern using a convolutional neural network for deep learning. Then, we conducted an evaluation experiment of the same method for the learning model of patterns and codes, and prove the applicability of the proposed method for the safety management of construction sites.

 The soundness of road pavements in our country is maintained by daily visual inspections as well as ro ad surface condition surveys once every several years conducted by road administrators. These inspection s and surveys require time-consuming work and expensive inspection costs, which are regarded as proble ms. To resolve the present situation, some methods have been proposed in recent years including a metho d for detecting road damage, such as cracks and potholes in the road pavement, using deep learning from i mages shot by a camera mounted on a traveling vehicle, and a method for calculating the percentage of cr acking. However, these methods are different from the rating method in actual inspections in some points.

 In this study, a method for rating cracking in the road pavement is developed by applying deep learning to video images from a drive recorder. It was found that the rating craking corresponding to the diagnostic categories I to III in the pavement inspection procedure can be carried out in consideration of cracks and patching by the devised method.

 During the construction work of mountain tunnels, monitoring the tunnel face is required all the time to prevent labor accidents. Small rock falls can be a sign of a large scale collapse, despite the time lag between them. However, it is difficult for the supervisor to detect small rock falls by visual monitoring. Therefore, in this research, a monitoring system was developed for detecting falling objects based on the background substraction. In order to monitor small rock falls, we limited the analysis scope of detecting falling objects process to the tunnel face which was detected by semantic segmentation and removed minute objects such as tunnel dust and video noise. Next, we conducted an image analysis by the developed system using footages which simulate falling rocks, the sign of a large scale collapse. The result showed that small rock falls could be detected, which implies the mitigation of labor accidents.

 In this paper, we propose a method of self-localization using acoustic reflection to introduce a robot into an indoor civil engineering environment. The acoustic sensor consists of a pair of horizontal omnidirectional loudspeakers and a horizontal omnidirectional microphone. The sound emitted from the loudspeaker is reflected from the wall and reaches the microphone. By measuring this time, the distance of the reflection path to each wall is measured. Using the distance of the reflection path, the rotation speed of the rotary encoder attached to the wheels of the robot, and the pre-supplied map information, we used an extended Kalman filter to estimate the self-position. Simulation verification showed that when the standard deviation of the acoustic sensor was 0.2 m, the positional error was less than 0.1 m. When the robot stage was moved 1 m in an anechoic room, the maximum error in position estimation was found to be ±0.05 m.

 The Ministry of Land, Infrastructure, Transport and Tourism (MLIT) mandates bridge managers to perform periodic inspections once every five years. Currently, sensors get smaller and cheaper, so many sensors are expected to be installed on bridges to monitor them. Humans judge the condition of the bridges from a large amount of sensing data and the situation data of sensor installation. In order to determine the status of a bridge by computers to be as well as humans, it is necessary to analyze not only the sensing data but also the relationship between the bridge and the installed sensors. Therefore, this study proposes a method for managing the relationship between bridges and sensors using databases such as relational databases and Not only SQL (NoSQL) databases. By managing the relationship between bridges and sensors with databases, the information on bridge properties can be managed with the sensing data. In the verification experiment, it was confirmed that the proposed method reduced the human work time by 30 % compared to the conventional information management based on drawings and documents.

 Improving working posture through quantitative evaluation is crucial because deterioration of working posture can cause occupational injuries including musculoskeletal disorders. Mechanical motion capture systems have been used in previous research on construction workers’ posture evaluation, but the deployment of them in practice could increase the project cost. Considering this concern, this paper proposes a method for estimating workers’ postures using Long Short-Term Memory (LSTM) and helmet-mounted terminals, which have been applied to construction sites. In the verification experiment, we confirmed that the proposed method was able to estimate the posture of a rebar worker with a large physical load by using the Ovako Working Posture Analyzing System (OWAS) method with a recall rate of over 80 %. This method can be used to quantify working posture and utilized in safety management at construction sites.

 In recent years, the use of tablet terminals has spread in many construction fieldworks. In this research, we focus on cameras and hand-written input functions standardized on such tablet terminals. This paper proposes a system that associates hand-written notes with AR (Augmented Reality) representation to enable smooth transmission and sharing of information on site. The operator's hand-written notes on the site are stored and automatically pop up in an AR expression when the mobile terminal captures the same scene with its camera. Conventional markerless AR systems using natural feature points have limitations in the observation angle because the reference points of image features are distributed on a plane. The proposed method prepares the reference points of natural features in 3D space using SfM (Structure from Motion). This paper reports the improved performance for AR display in viewing angle and wider varieties of objects' shapes. We also validated the proposed system's effectivity by the experiments using actual different kinds of pillar-shaped items for AR display hand-written notes, assuming concrete structures' inspection work.

 This paper presents the development of AR visualization systems for underground structure using ARKit. We developed two AR systems: a system to superimpose the CAD model of the underground buried structure on the design drawing and a system to superimpose the CAD model on site. In the system that superimposes the CAD model on the design drawing, we tried to superimpose the city model at the same time in order to easily grasp the positional relationship. In addition, in the system for superimposing CAD models in the field, we examined the masking processing method and the method for improving the accuracy of the superimposition position of the CAD model. In order to verify the effectiveness of this system, The present systems are applied to an actual example of underground structure.

 Crack extraction using deep learning for patches cut out from road surface images has the problem of omission or over-extraction for areas other than general dense-graded pavement such as drainage and concrete pavement. Therefore, we proposed and verified a two-stage road surface crack extraction method that classifies road surfaces into four classes in the previous stage and extracts cracks using the results and road surface patch images in the next stage. As a result, improvements of 8.3% to 0.6% in F value and 0.14 to 0.08 in AUC ware observed compared to the case where the road surface type was not used.

 Furthermore, as a result of proposing a method for calculating the crack rate based on the mesh method of the standard pavement survey / test method manual from the crack extraction results for such patch images, the correlation with the visual analysis in the 20m unit length evaluation crack rate was 0.94.

 There is still a relatively high incidence of work accidents during the mountain tunnel construction when compared to the incidence rates in the other construction industry.In the tunnel construction, it is important to monitor the deformation of tunnel face in order to prevent the accidents. Especially, the workers are injured due to the rock fall events at a work site.Therefore, the authors proposes the method to detect the fall of rock by using a depth sensor and image processing in the paper. The experimental results show the effectiveness of the proposed method.

 In Japan, there are many disasters such as typhoons, earthquakes, and tsunamis, and it has been pointed out that there is a danger of the collapse of block walls. Therefore, there is an urgent need to understand and inspect the current state of block walls. Tokushima University is conducting a field survey based on the inspection indicators of the Ministry of Land, Infrastructure, Transport and Tourism with the cooperation of prefectures and municipalities. However, since block walls are scattered in various places, there is a limit to covering a wide area in a short time in the current field survey. On the other hand, construction consultants and surveying companies are making increasing use of sensing data such as lasers and cameras. Therefore, in this research, we will refer to the method of improving the maintenance of block walls using sensing data, and propose a technology to automatically extract block walls from point cloud data.

 As the population declines, it is expected that the scale of activity in the urban will shrink. In such the situation, it is effective to analyze and evaluate using an economic model that can express space in order to consider the development of infrastructure that exists in the urban for a long period of time. On the other hand, in recent years, as changes in the conditions for building the model, there are enhancement of the data environment and complexity of setting. Against this background, the computational load of the model tends to increase, and reducing it is one of the issues. This paper aims to show the improving the calculation time of the economic model that can express the space. We applied RMSprop and Adam, which are used in deep learning, to the optimization calculation of the existing model, and compared the results. As a result, it was verified that RMSprop and Adam contributed sufficiently to the efficiency of calculation time depending on the setting of the learning rate.

 There are 530,000 small bridges with lengths of 2 - 15 m. An ordinance by the Ministry of Land, Infrastructure, Transport and Tourism requires periodic close-up visual inspection of bridges once every 5 years. Significant time was spent identifying the bridges to be inspected. The aim of this study is to improve the inspection efficiency for small bridges whose names cannot be confirmed on-site by using a tablet terminal to display names and overviews of target bridges. The system also eliminates inspection errors and omissions, improving the reliability of inspection results. The developed system uses RFID and QR codes as a tool for reading bridge names with a tablet, and we demonstrated its on-site use. We selected as study sites five small bridges in Niigata City. RFID tags can be read from 23 - 27 cm, so bridge sides are appropriate locations for their attachment. Regarding system operability, a bridge maintenance expert described the system as having mainly simple operations, making it easy to use and not burdensome.

 In order to prevent fatal accidents caused by falling concrete skin and rocks in tunnel construction, authors developed of tunnel face monitoring system applying a millimeter-wave, 79GHz high-speed imaging radar. The radar is able to detect the minute deformation and vibration of tunnel face in non-contact.

 In this paper, authors considered the precursors of face and rock fall as minute movements of the surface and changes in frequency charecteristis due to drilling work, etc., and conducted fundamental experiment using concrete panel test pieces to detect that would be captured by the radar.

 Performance and availability were confirmed through a fundamental experiment and a distributed vibration measurement. Also, authors described the problem in coordinate transformation that coordinate from the measured value obtained by the radar to the face when applying the radar to the tunnel face, propose a simple conversion method as a countermeasure.

 The trial measurement with the simple conversion method in the actual tunnel is conducted, and the result is confirmed.

 Currently, as an efficient inspection method for civil engineering structures, The Mobile Mapping System (MMS) is used for this purpose. In addition to the conventional TOF method, the laser scanners installed in MMS include, In recent years, the phase detection method (PS method) has been adopted. In previous studies, the basic accuracy of MMS has been verified using the PS method, but There has not been sufficient verification focusing on the driving speed. In this study, MMS measurements by the PS system were made at speeds of 20 km/h, 40 km/h, 60 km/h, and 80 km/h, and the measurement accuracy was compared to that of the other methods. We compared them. As a result, the thickness calculation results show that the TOF method, the mean value of the calibration increased with the increase in speed, but with the PS method, there is no significant change in the RMS error of the calibration as the speed is increased. It was shown that the measurement accuracy was stable and good.

 Positioning technology using a positioning satellite system is required for the development of next-generation ITS technologies such as autonomous driving. The number of positioning satellites that can be captured is small in an environment where skyscrapers are lined up or the sky is covered with vegetation and signals affected by multipath such as reflected waves and diffracted waves, greatly reduce the positioning accuracy of satellite positioning systems. In this study, the difference in signal strength was obtained for each vehicle behavior whether the vehicle was running or stopped, and the effect of selecting a satellite suitable for positioning calculation was verified. As a result, it was confirmed that the positioning rate and the fix rate were improved depending on the driving direction, driving environment, and vehicle behavior.

 Mobile Mappinng System(MMS) is used for creating road ledgers and inspectong structures. In previous studies institutional verification was condecuted targeting the road surface snd wall surface of MMS. However, accuracy verification for the upper part such as the ceiling has not been sufficiently performed. Therefore, in this study, that measured MMS on the upper part and on the road surface. and this measurement was done at 2 rates, that laser shots is million points per second and fifty thousand points per second. As a result, when the target is installed the upper space and on the road surface, that measurment accuracy of the upper space was good.

 Since Japan has an urgent need to respond to a super-aging society in recent years, the number of wheelchair users increases, and the government promotes barrier-free access. However, available barrier information around wheelchair users' lives is not sufficiently detailed. Therefore, this research proposes a barrier verification and mapping system that is easy to use and easy to understand for facility managers to maintain and manage barrier-free facilities. As a depth camera acquires a 3D point cloud of the target space, interference with volume could be occupied by the wheelchair is found as physical barriers. The detected barriers are highlighted and integrated with a wide-angle color camera frame to produce an augmented reality representation for generic users' easy understanding. Besides, collected barrier information can be efficiently reused for 3D mapping and accumulated through an image-based 3D reconstruction technique, SfM (structure from motion). This paper reports the implementation method and the barrier verification result obtained by the experiment. The experimental results show that the proposed system provides the compatibility of real-time barrier verification and efficient information recording.

 Mobile Mapping System (MMS) is used for maintenance of civil infrastructures such as roads and rivers, and generation of high precision three-dimensional map for autonomous driving. However, it is difficult for all local govenments to use MMS because it is very expensive. Though we designed vehicle-mounted sensing unit with commercially available and non-expensive sensors, the unit had problems in the unit shaking and sensors installation position. In this research, new unit designed and constructed for generating high-precision three-dimensional spatial data. It solves problems of the existing unit. And, we devised analyzing method for generating three-dimensional spatial data by proposed unit. Through measurement experiments, we confirmed that the proposed unit can generate high-density point cloud data within the catalog specifications of laser scanner and GNSS.

 In this study, the study area is selected the regions that happened slope failure accompanied by topographical changes caused by Northern Kyushu heavy rain in July 2017 and take ALOS-2/PALSAR-2 data is used as the satellite information. Under the same conditions, the results show that firstly, single-polarization backscatter differential analysis ex-tracts the terrain changes more accurately than the coherence analysis. Secondly, based on the slope angle, aspect angle and local incident angle, we also classify the slope failure that is able to extracted, incorrectly extracted and unextracted. As a result, slope angle less than 15° can be extracted well by the backscatter differential analysis with almost no error. Finally, the aspect angle is well extracted in the east-west direction using the coherence analysis. In contrary, the south direction is well extracted by the backscatter differential analysis. Furthermore, the larger of off nadia angle, the easier to extract the areas where the local incident angle is small.