Light detection and ranging (LiDAR) generating point clouds is rapidly expected to apply to the road monitoring. However, existing methods for co-registration of two sets of point clouds, such as Iterative Closest Point (ICP) method, are not robust against the occlusion caused by pedestrians or vehicles. Therefore, this paper presents a method that achieves accurate co-registration of the point clouds even though the data are occluded. The proposed method composes of two processing. The first processing takes step-by-step correction for a rough co-registration by utilizing normal derived from planes vertically standing, e.g. building wall surfaces and. After the rough correction is completed, the second processing utilizing ICP method is implemented for an accurate co-registration. In the experiments, we used the point clouds observed by using terrestrial LiDAR data, and the parts of the point clouds were excluded, representing the occlusion by objects. The validation results show that the proposed method achieved a high accuracy even for the occluded data.

View full abstract

 Restoration work following a sediment disaster or volcano disaster may be executed in extremely dangerous places. Construction in such dangerous places can be executed safely if it can be done using construction equipment that can be operated remotely while observing camera images. Construction using such remote control type construction machinery is called “unmanned construction”, and it has been carried out more than 150 sites in disaster prone Japan. It is said that unmanned construction is generally low efficiency construction so an improvement of efficiency have been needed.
 If an operator good at unmanned construction and an operator NOT good at unmanned construction exist, it is possible to distinguish operator degree of achievement easily by investigating the characteristic of both operator. And, if we select only a suitable operator and dispatch them to the construction site, it is possible to improve construction efficiency of the unmanned construction site.
 So in this study, first, we distinguished both operator by the experiment on test field. Next, the characteristic survey of both operators are done using the eye-tracking camera. Finally, we suggested the method for distinguishing the degree of achievement of the operator easily and for selecting a suitable operator for unmanned construction.

View full abstract

 Recently, robots utilizing UAV (Unmanned Aerial Vehicle) or the like for next generation social infrastructure are developed, and acquisition of images for inspection of concrete structures is becoming more efficient. It is also expected to develop an effective image processing system against large amount of data, to maintain and manage an enormous number of structures efficiently. In this paper, we propose an image processing method for reliable evaluation, and it realizes efficient visual inspection of concrete structures. At first, we propose a semi-automatic crack extraction method in which an operator roughly specifies the position of cracks on the image. We also propose applying super-resolution and an evaluation method of crack width using a crack scale. Furthermore, we propose an image synthesis method that corrects lens distortion aberration using only the straight line on the image and interpolates occlusion using multiple images from different viewpoints. Experimental results show that cracks can be extracted with high accuracy by the semi-automatic processing even in low spatial-resolution and low contrast images. And it also suppresses the calculation cost and the photographing cost, compared to the conventional method.

View full abstract

 Precise deterioration prediction of individual section is effective for planning and implementing effective maintenance and management measures of social overhead capital, but deterioration prediction of social capital is difficult because complicated factors are involved. This paper proposes a method of predicting the road pavement condition values by introducing ADAM and dropout into multi-layer perceptron (MLP) of neural network and recurrent neural network (RNN) that can handle time series data. The proposed method was applied to the prediction of the pavement condition survey values consisting of cracking ratio, rutting depth and roughness three years later and was evaluated by RMSE representing prediction error. Comparing the MLP and the linear regression model that are the most widely and conventionally used for predicting pavement condition survey values for individual sections of pavement, the RMSE of the crack rate was decreased by approximately 13%. By using the RNN model, we also found that the RMSE decreases about 7% with respect to MLP.

View full abstract

 The utilization of the Synthetic Aperture Radar (SAR) has wide application areas recently, and one of them is to grasp the situation of damaged area after the occurrence of a disaster. The small ground displacement is obtained, for example, by Differential Interferometric SAR (DInSAR) technique. However, there are few researches for performing the feature of the errors of DEM, and the characteristics and the errors of fringes of the deformation were not revealed yet. We examined the method to evaluate the phase noise utilizing the errors of DEM of fixed points. In the current study, we evaluated the errors of DEMs focusing on the flat ground targets such as roads, parking lots, and parks. The results revealed the relationships between errors and baseline using multiple pairs of images with different frequency, acquisition mode, and the acquisition direction.

View full abstract

 Evacuation information on the tsunami inundation forecast hazard map is required to be up to date. However, the hazard map can not be updated frequently due to the human cost for basic research, and the calculation cost for high precision simulation. Therefore, we classify land use by combining texture analysis with local fractal dimension, DEM data, and coastline data. Based on this result, we forecast inundation of the tsunami with small calculation cost by energy conservation method. Furthermore, we construct the road network from road edge data using image processing, and search the evacuation routes with the Dijkstra method considering horizontal distance, elevation, and inundation depth. Using these method, we create a tsunami inundation forecast hazard map which is easily updated.

View full abstract

 This paper presents a retrieval method of similar inspection records based on experienced inspectors' evaluation in order to realize more efficient maintenance inspection for road structures such as bridges and tunnels. The proposed method introduces distance metric learning based on experienced inspectors' evaluation, and enables calculation of distances suitable for the retrieval of inspection records. Furthermore, the proposed method performs rank-level fusion that can successfully integrate retrieval results obtained from different aspects, i.e., image and text information. Consequently, inspectors can evaluate distresses referring to past similar records by our retrieval method. Therefore, since it is possible to reduce the differences of evaluation between inspectors for distress assessment, more efficient and accurate maintenance inspection becomes feasible. The effectiveness of the proposed method is verified by several experiments with actual inspection records.

View full abstract