Construction management system and construction management method

Abstract

To provide a construction management system and a construction management method capable of quantitatively judging a construction quality in a scattering process.SOLUTION: A construction management system 1 managing a construction in a scattering process comprises: a capturing unit 12 capturing a scattered materials for each area; and a processing unit 20 obtaining information regarding construction quality control by analyzing the captured images. The processing unit 20 includes: a constituent detection portion detecting constituents of the material in the captured image; and a particle size distribution estimation portion estimating a particle size distribution of the material from the detected constituents. The construction management system 1 further comprises a moving portion 10 capable of mounting the capturing unit 12. The capturing unit 12 may capture the material while moving on the scattered surface.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a construction management system and a construction management method.

Background Art

[0002] In the process of manufacturing concrete or CSG (Cemented Sand and Gravel) in the construction of dams and the like, batch plants, CSG mixing plants, etc. are used. In these plants, it is required in quality control that the particle size of locally generated materials (aggregates, earth and sand, etc.) used is within a predetermined range (see Fig. 14). For quality control of CSG, it is necessary to grasp the particle size distribution of locally generated materials. The particle size distribution is managed, for example, in five grades (0 - 5 mm, 5 - 10 mm, 10 - 20 mm, 20 - 40 mm, 40 - 80 mm). Conventionally, quality control tests have been carried out manually, and samples are intermittently sampled at a fixed frequency (once per hour at the initial stage of construction) for particle size measurement (for example, see JIS A 1102 "Sieving test method for aggregates"). The quality control test by manpower is, for example, to sieve the sample by particle size with a sieve and dry it, and measure the weight of such a sample to grasp the particle size distribution. Also, as a technique for grasping the particle size distribution, a technique has been proposed to photograph a material (for example, locally generated material) and measure the particle size distribution of the material in real time based on the photographed result (for example, see Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] It is generally known that material segregation during embankment construction is a problem when using materials with a large maximum particle size. For example, in the embankment construction of rockfill dams, coarse-grained materials with a maximum particle size of 500 mm or more are used, resulting in uneven particle sizes. This tends to cause large chunks to concentrate on the outer edges of the embankment surface during unloading and spreading. Traditionally, workers visually checked for material unevenness, but this relied on the workers' subjective judgment, which meant that accurate quality control was not possible. Furthermore, visual inspection required a great deal of effort to check a vast area. Even if material quality control is performed by sampling when the material is extracted from the quarry, material segregation problems can still occur, and similar problems can occur in construction projects that use materials containing components of different sizes, not just rockfill dams. From this perspective, the present invention provides a construction management system and a construction management method that can quantitatively determine the quality of the spreading process. [Means for solving the problem]

[0005] The construction management system according to the present invention is a construction management system that manages the construction of the spreading process. This construction management system includes a shooting unit that photographs the spread material area by area, and analyzes the captured images to obtain information regarding the quality control of the construction. by area The processing unit comprises a component detection unit that detects the components of the material in the captured image, and a particle size distribution estimation unit that estimates the particle size distribution of the material from the detected components. A particle size distribution determination unit that determines whether the particle size distribution of each area falls within a predetermined control range, It has. In the construction management system according to the present invention, the particle size distribution is estimated by photographing the material after it has been spread, area by area. The particle size distribution estimated for each area can serve as an indicator of the bias in the placement of materials in each area. Therefore, the bias of the constituent materials can be quantitatively judged based on the particle size distribution estimated for each area. As a result, there is no room for human subjective judgment, and construction management can be performed more accurately.

[0006] The construction management system may further include a mobile unit on which the imaging unit can be mounted. The imaging unit moves along the surface from which the material is being spread and takes photographs of the material. The moving unit may be an unmanned aerial vehicle and may move at a predetermined speed while maintaining a constant distance to the dispensing surface. In this case, the imaging unit will photograph the dispensing surface at predetermined time intervals. Furthermore, the moving unit may be a construction machine used in the spreading process or a construction machine used in the compaction process performed after the spreading process. In that case, the imaging unit will photograph the spreading surface during the spreading process or the compaction process. This approach makes it possible to easily manage construction over a vast area.

[0007] The component detection unit individually detects the components captured in the captured image using a trained model that has learned the components using a learning model that corresponds to instance segmentation. This method allows for the separate detection of overlapping and contacting components at the pixel level. Therefore, more accurate construction management is possible. The trained model may be machine-learned to detect the components by inputting segmented images obtained by dividing the captured image. This method increases the apparent size of the materials in the captured image, even when the distance between the camera and the surface from which the material is being dispensed is large, making it easier to recognize the components. As a result, more accurate construction management becomes possible.

[0008] The construction management method according to the present invention is a construction management method for managing the construction of the spreading process. This construction management method includes a photographic process in which the spread material is photographed area by area, and by analyzing the photographed images, information regarding the quality control of the construction is obtained. by area The process includes a required processing step. The processing step includes a component detection step for detecting components of the material in the captured image, and a particle size distribution estimation step for estimating the particle size distribution of the material from the detected components. , a particle size distribution determination step that determines whether the particle size distribution of each area falls within a predetermined control range, It has. In the construction management method according to the present invention, the particle size distribution is estimated by photographing the material after it has been spread in each area. The particle size distribution estimated for each area can serve as an indicator of the bias in the placement of materials in each area. Therefore, the bias of the constituent materials can be quantitatively judged based on the particle size distribution estimated for each area. As a result, there is no room for human subjective judgment, and more accurate construction management is possible. [Effects of the Invention]

[0009] According to the present invention, the quality of the application process can be quantitatively determined. [Brief explanation of the drawing]

[0010] [Figure 1] This is a schematic diagram of the construction management system according to the first embodiment of the present invention. [Figure 2] This figure shows an application scenario of the construction management system according to the first embodiment of the present invention. [Figure 3] This is a functional configuration diagram of a construction management system according to the first embodiment of the present invention. [Figure 4] This is a functional configuration diagram of the analysis unit. [Figure 5] This is an example of a photograph taken by the photography department. [Figure 6] This is an image of dividing a captured image. [Figure 7] This is a detection result image that reflects the detection results in the captured image. [Figure 8] This is an image showing the detection result when object recognition is performed on a captured image without splitting it. [Figure 9] This is an example flowchart illustrating the process in calibration mode. [Figure 10] This is an example flowchart illustrating the process in learning mode. [Figure 11] This is an example flowchart illustrating the processing in the particle size distribution estimation mode. [Figure 12]This is a schematic configuration diagram of the construction management system according to the second embodiment of the present invention. [Figure 13] This is a schematic configuration diagram of the construction management system according to the third embodiment of the present invention. [Figure 14] This is an example of information used for quality control of locally produced materials.

Embodiments for Carrying Out the Invention

[0011] Hereinafter, embodiments for carrying out the present invention will be described in detail with appropriate reference to the drawings. Each figure only schematically shows the present invention to such an extent that it can be sufficiently understood. Therefore, the present invention is not limited only to the illustrated examples. In each figure, common components and similar components are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

[0012] [First Embodiment] <Regarding the configuration and application scenarios of the construction management system according to the first embodiment> Referring to FIGS. 1 and 2, the configuration and application scenarios of the construction management system 1 according to the first embodiment will be described. FIG. 1 is a schematic configuration diagram of the construction management system 1 according to the first embodiment. FIG. 2 is a diagram showing the application scenarios of the construction management system 1 according to the first embodiment. The construction management system 1 shown in FIG. 1 is a system for managing the construction of the spreading process. The material to be spread includes granular constituents (for example, stones). The size of the constituents is not particularly limited, and may be, for example, from several millimeters to several meters. Construction Management System 1 can be used in various construction projects that require a spreading process, and the types and uses of construction projects to be managed are not particularly limited. In this embodiment, it is assumed that Construction Management System 1 will be used in the construction of a rockfill dam embankment. An example of the process for construction of a rockfill dam embankment is shown in Figure 2. Construction of a rockfill dam embankment is carried out in the following order, for example, "material extraction process" from a quarry, "transportation process" of the extracted materials, "unloading process" of the transported materials, "spreading process" of the materials, and "compaction process". Construction Management System 1 is implemented between the "spreading process" and the "compaction process".

[0013] As shown in Figure 1, the construction management system 1 comprises an imaging unit 12 and a processing unit 20. The imaging unit 12 is preferably mounted on a movable unit 10 that can move on the spreading surface. The mobile unit 10 is movable on the spreading surface. In this embodiment, the mobile unit 10 is an unmanned aerial vehicle (UAV), and the camera unit 12 photographs the spreading surface from above. For example, a worker operates the control unit 15 to fly the mobile unit 10 and presses the camera button on the control unit 15, causing the camera unit 12 to photograph the spreading surface. The captured image is transmitted to the processing unit 20 via the relay unit 30. The mobile unit 10 may also be a vehicle that travels on the spreading surface (for example, construction machinery). The mobile unit 10 preferably has an intelligent flight mode. In intelligent flight mode, for example, by pre-setting the flight altitude, flight speed, flight path, etc., automatic flight is possible with only the press of a button to start the flight. The intelligent flight mode preferably includes control of the timing of the imaging unit 12's imaging, in which case it is set to take images at predetermined time intervals, for example. By using intelligent flight mode, the operator can automatically photograph the entire surface being sprayed by simply checking the flight status. In this embodiment, imaging will be performed using intelligent flight mode.

[0014] The imaging unit 12 shown in Figure 1 is used to photograph the material being spread during the spreading process. By performing imaging in intelligent flight mode, the imaging unit 12 can photograph the spread material area by area. An area is the smallest unit for quality control of the construction, and at least one photograph is taken for each area. Multiple photographs may be taken for a single area, or the shooting range of the imaging unit 12 in a single photograph may be defined as an area. Areas may be set without gaps on the spreading surface, or there may be gaps between adjacent areas. The imaging unit 12 is installed, for example, on the underside of the flying mobile unit 10, with the optical axis of the lens pointing vertically downward. The imaging unit 12 is preferably a high-resolution digital camera or a video camera capable of acquiring digital images. The processing unit 20 is a computer consisting of a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), etc. The processing unit 20 obtains information regarding construction quality control by analyzing the images captured by the imaging unit 12. Quality control here refers to ensuring that there is no bias in the arrangement of components contained in the spread material, and this quality control is important for ensuring the required strength of the structure. In this embodiment, the presence or absence of bias in the material is determined using the particle size distribution. Therefore, the information regarding construction quality control is the particle size distribution for each area of ​​the spreading surface.

[0015] Referring to Figure 3, the more specific configuration of the construction management system 1 will be explained. Figure 3 is a functional configuration diagram of the construction management system 1. The mobile unit 10 includes an IMU (Inertial Measurement Unit) 11, an imaging unit 12, a GPS (Global Positioning System) 13, and a communication unit 14. The IMU11 has sensors for various physical information (e.g., an accelerometer, an angular velocity (gyroscope) sensor, etc.) and can detect, for example, three-dimensional inertial motion (translational and rotational motion in the orthogonal three-axis directions). The information detected by the IMU11 is used for attitude control of the moving unit 10. The imaging unit 12 photographs the sprinkling surface area by area. The images captured by the imaging unit 12 and the focal length at the time of shooting are sent to the processing unit 20 for analysis. The GPS 13 calculates the position of the mobile unit 10 (e.g., latitude, longitude, altitude) using orbital position information and time information received from positioning satellites such as GPS satellites. Other means of acquiring position information may be used instead of the GPS 13. The position information obtained by the GPS 13 is sent to the processing unit 20 in association with the captured image. The communication unit 14 enables communication with the processing unit 20. The communication unit 14 performs wireless communication, such as Wi-Fi.

[0016] As shown in Figure 3, the processing unit 20 comprises a communication unit 21, an analysis unit 22, and a notification unit 23. The communication unit 21 enables communication with the processing unit 20. It is connected to the relay unit 30 via a LAN (Local Area Network). The relay unit 30 acts as a relay point (access point) for wireless communication. The processing unit 20 acquires information from the mobile unit 10 via the communication unit 21. The acquired information is input to the analysis unit 22. The analysis unit 22 learns the components of the material 9 used for dispensing using machine learning and detects the components in the images captured by the imaging unit 12. Then, the analysis unit 22 calculates the size (particle size) of each component based on the detection results and determines the particle size distribution of the material in the captured images. The analysis unit 22 calculates the particle size distribution for images taken at multiple locations and determines whether the particle size distribution of each area falls within a predetermined control range. If the particle size distribution does not fall within the predetermined control range in an area, it can be determined that there is a bias in the components, and countermeasures such as re-applying the material are necessary in that area. The analysis unit 22 passes the analysis results, such as the particle size distribution, to the notification unit 23.

[0017] The notification unit 23 notifies workers and managers of the analysis results performed by the analysis unit 22. The method of notification of the analysis results by the notification unit 23 is not limited and may include image display, audio output, email notification, etc. For example, if there is an area where the particle size distribution does not fall within a predetermined control range, the notification unit 23 will notify that there is an area with construction defects, along with information about that area (e.g., latitude and longitude at the time of photography). The analysis unit 22 may calculate the particle size distribution for the images taken at all locations, and estimate the particle size distribution of the entire spreading surface by statistically processing the results (e.g., averaging), and determine whether the particle size distribution falls within a predetermined control range. This makes it possible to control the quality of the spread material itself.

[0018] Next, the processing of the analysis unit 22 will be explained with reference to Figure 4. Figure 4 is a functional configuration diagram of the analysis unit 22. In this embodiment, it is assumed that the analysis unit 22 detects components from captured images using deep learning, one of the machine learning methods, and the case in which MASK R-CNN (Regions with Convolutional Neural Network) that supports instance segmentation is used will be explained. Here, segmentation outputs which class each pixel in an image belongs to, while instance segmentation further divides the region for each individual object (extracting the object region of each individual object). Therefore, instance segmentation makes it possible to separately detect overlapping or touching objects of the same class on a pixel-by-pixel basis. MASK R-CNN is an extension of Faster R-CNN. Since the processing of MASK R-CNN is publicly known, a detailed explanation will be omitted. MASK R-CNN is described in the following literature, for example. Kaiming He et al., "Mask R-CNN", Internet<https: / / arxiv.org / pdf / 1703.06870.pdf>

[0019] Thus, because MASK R-CNN supports instance segmentation, it can determine whether or not a pixel is a component to be detected. The analysis unit 22 then counts the pixels determined to be components in the captured image and calculates the area and size (particle size) of each component, and calculates the particle size distribution based on the area corresponding to the particle size category. MASK R-CNN supports the detection of multiple types of objects (classification of multiple classes), and can recognize the type (class) of each object when multiple types of objects are captured in a single image. In this embodiment, a mixed material is assumed as the target object for measuring the particle size distribution, and the case of detecting coarse-grained materials such as aggregates is described. In other words, only the "coarse-grained material class" related to coarse-grained materials is detected, and fine-grained materials such as soil and sand are not detected. Note that the class classification may be other than that described in this embodiment.

[0020] As shown in Figure 4, the analysis unit 22 operates in three modes: "calibration mode," "learning mode," and "particle size distribution estimation mode," and has the functions necessary for these operations. The calibration mode is a mode for making adjustments according to the conditions under which the imaging unit 12 is capturing images. The learning mode is a mode for learning the relationship between objects captured in the captured image and the components of the material to be detected. The analysis unit 22 performs learning using the captured images taken by the imaging unit 12. The particle size distribution estimation mode is a mode for detecting the components of the material captured in the captured image and estimating the particle size distribution of the material based on the detection results. The analysis unit 22 performs particle size distribution estimation using the captured images taken by the imaging unit 12. The calibration mode is performed before the learning mode and particle size distribution estimation mode. The particle size distribution estimation mode is performed after the learning mode has been run. After the adjustment in calibration mode is completed, the distance between the imaging unit 12 and the spreading surface is not changed. In this embodiment, the mobile unit 10 is controlled using intelligent flight mode, and the flight altitude of the mobile unit 10 is always constant (for example, "15m"). Therefore, the distance between the imaging unit 12 and the spreading surface is the same in all three modes: "calibration mode," "learning mode," and "particle size distribution estimation mode."

[0021] The analysis unit 22 includes a calibration image input unit 41, a calibration parameter generation unit 42, and a storage unit 43. These functions are primarily related to the "calibration mode". Furthermore, the analysis unit 22 includes a learning image input unit 51, a learning image segmentation unit 52, a training data generation unit 53, a deep learning unit 54, and a storage unit 55. These functions are mainly related to the "learning mode". Furthermore, the analysis unit 22 includes an estimation image input unit 61, an estimation image splitting unit 62, a component detection unit 63, a recognized image merging unit 64, an area calculation unit 65, a diameter determination unit 66, a mass conversion unit 67, a particle size distribution estimation unit 68, and a particle size distribution confirmation alarm unit 69. These functions are mainly related to the "particle size distribution estimation mode". Here, Figure 5 shows an image of the captured image 8 taken by the imaging unit 12. As shown in Figure 5, the captured image 8 shows the material 9 that was spread out in the spreading process. The material 9 contains granular components 9a.

[0022] (Functions related to calibration mode) The calibration image input unit 41 has a function for inputting an image to be used for calibration (calibration image "correction image") from the imaging unit 12. The calibration parameter generation unit 42 has a function for determining calibration parameters from the input calibration image (corrected image). The calibration parameters include, for example, information necessary for image correction and coefficients that show the relationship between real space length and pixels in the captured image. The coefficients that show the relationship between real space length and pixels in the captured image are determined by the sensor size of the imaging unit 12, the focal length, and the distance between the imaging unit 12 and the sprinkling surface. The calibration parameters are stored in the storage unit 43 and are used, for example, to determine the diameter of the constituent 9a when estimating the particle size distribution.

[0023] (Functions related to learning mode) In learning mode, the system learns to perform object detection (including location determination) and object recognition from captured image 8. The learning image input unit 51 has the function of inputting captured images 8 (learning images) to be used for learning from the shooting unit 12. The learning images are preferably captured under the same conditions as when estimating the particle size distribution, for example, images of the spreading surface (material 9) while the mobile unit 10 is flying using intelligent flight mode. The learning image splitting unit 52 has the function of splitting the captured image 8 (learning image) used for training. Figure 6 shows an image of the splitting of the captured image 8. Note that the splitting of the captured image 8 includes the process of dividing the captured image 8 into multiple regions. By splitting the captured image 8 into separate split images 8a, the apparent size of the components 9a in the captured image 8 increases, making it easier to recognize the components 9a. Therefore, splitting the captured image 8 is effective, for example, when the distance between the shooting unit 12 and the dispensing surface is large.

[0024] The training data generation unit 53 has a function for generating training data. The training data generation unit 53 displays the segmented images 8a obtained from the learning image segmentation unit 52 on the screen and accepts the setting of tags (for example, information indicating that it is a component 9a) by a human (for example, a worker). The training data generation unit 53 creates training data by tagging the segmented images 8a with information such as the following. • Information on the frame (polyline) surrounding the aggregate, which is component 9a. • Information regarding the class of aggregate, which is component 9a. • Information regarding the mask of the aggregate component 9a The training data generation unit 53 generates training data by associating the segmented images 8a with tag information, and outputs the generated training data to the deep learning unit 54. Alternatively, training data may be created using the original captured image 8 instead of the segmented images 8a. The deep learning unit 54 has the function of training a learning model (network). In this embodiment, the deep learning unit 54 trains MASK R-CNN using training data created by the training data generation unit 53. The deep learning unit 54 trains the learning model to detect components 9a (aggregates, etc.) contained in the material 9 by inputting the captured image 8 or its segmented image 8a of the material 9 captured by the imaging unit 12. The trained model (trained model) after training is completed is stored in the storage unit 55.

[0025] (Functions related to particle size distribution estimation mode) In particle size distribution estimation mode, object detection (including location determination) and object recognition are performed from the captured image 8 of material 9, and the particle size distribution of the constituent 9a is estimated from the results. The estimation image input unit 61 has the function of inputting captured images 8 (estimation images) used for estimating the particle size distribution from the capture unit 12. The estimation images are preferably captured under the same conditions as the training images used in training, for example, images of the spreading surface (material 9) while the mobile unit 10 is flying using intelligent flight mode. The estimation image segmentation unit 62 has the function of segmenting the captured image 8 (estimated image) used for estimation. The method of segmenting the captured image 8 is the same as that of the learning image segmentation unit 52, and the segmentation of the captured image 8 includes the process of dividing the captured image 8 into multiple regions.

[0026] The component detection unit 63 has the function of detecting components 9a (aggregates, etc.) contained in the material 9 from the captured image 8 or its segmented image 8a captured by the imaging unit 12. In this embodiment, the component detection unit 63 detects components 9a (aggregates, etc.) using a trained model trained by the deep learning unit 54. The component detection unit 63 inputs the segmented image 8a obtained by dividing the estimation image into the trained model and obtains the detection result (information on the detected components 9a (aggregates, etc.)) output from the trained model. The recognition image merging unit 64 has the function of merging the segmented images 8a after detecting the components 9a (aggregates, etc.) to return them to the size of the original captured image 8. By merging the segmented images 8a to return them to the size of the original captured image 8, the size corresponds to the calibration parameters obtained in calibration mode. Note that if calibration parameters are obtained using the segmented images 8a, it is also possible to perform subsequent processing on each segmented image 8a without merging the segmented images 8a after detecting the components 9a (aggregates, etc.).

[0027] Figure 7 shows the detection result image, which reflects the detection results in the captured image 8. In Figure 7, detected objects (components 9a (aggregates, etc.)) are highlighted by adding hatching, etc. As shown in Figure 7, even if multiple components 9a overlap or are in contact, they are detected as different objects. Figure 8 also shows the detection result image when object recognition is performed without dividing the captured image 8. In Figure 8, smaller components 9a are not recognized compared to Figure 7. The reason why the recognition rate of small components 9a is good in Figure 7 is that by dividing the captured image 8 into separate divided images 8a, the apparent size of the components 9a in the captured image 8 increases, making it easier to recognize the components 9a.

[0028] The area calculation unit 65 has the function of calculating the area of ​​the components 9a (aggregates, etc.) detected by the component detection unit 63. The area calculation unit 65 calculates the area (number of pixels) of each component 9a in the captured image 8 and outputs the calculation result to the diameter determination unit 66 and the mass conversion unit 67. The diameter determination unit 66 has the function of determining the diameter of the component 9a (aggregate, etc.) detected by the component detection unit 63. The diameter determination unit 66 adds diameter information to the detected component 9a information and outputs it to the mass conversion unit 67. The mass conversion unit 67 has a function for calculating a mass conversion coefficient (also called a "weight conversion coefficient") for each particle size category. The mass conversion coefficient is the mass per predetermined area of ​​the constituent material 9a (aggregate, etc.) in the captured image 8. The mass conversion unit 67 has known values ​​for calculating the mass conversion coefficient by some method (for example, pre-registration by an operator). The mass conversion unit 67 calculates the mass conversion coefficient for each particle size category based on the total area of ​​each particle size category of the constituent material 9a (aggregate, etc.) and the known values ​​for calculating the mass conversion coefficient. The mass conversion unit 67 outputs the calculated mass conversion coefficient to the particle size distribution estimation unit 68. The mass conversion coefficient may be calibrated according to the water content (surface water content) of the aggregate, etc. The higher the water content, the larger the aggregate particle size. That is, the higher the water content, the more fine-grained material adheres to the coarse-grained material and forms aggregates, so the apparent particle size of the aggregate becomes larger.

[0029] The particle size distribution estimation unit 68 calculates the particle size distribution of the material 9 based on the total area for each particle size category of the constituent material 9a (aggregate, etc.) in each captured image 8 and the mass conversion coefficient. The particle size distribution estimation unit 68 obtains the particle size distribution for each captured image 8 from the captured image 8. The particle size distribution obtained from each captured image 8 serves as an indicator of the bias of the material 9 placed in each area. If multiple images are taken in one area, the particle size distribution estimation unit 68 may obtain the particle size distribution of that area by statistically processing these results (for example, averaging them). The particle size distribution estimation unit 68 displays the calculated particle size distribution for each area on a display unit (not shown), for example. The particle size distribution confirmation alarm unit 69 determines whether the particle size distribution in each area falls within a predetermined control range, and if there is an area that does not fall within the predetermined control range, it sends an alarm via the notification unit 23. Since an area where the particle size distribution does not fall within the predetermined control range can be judged to have an imbalance in the constituent material 9a, the worker who receives the notification takes appropriate action, such as reworking, in that area.

[0030] <Regarding the construction management method of the construction management system according to the first embodiment> The construction management method of the construction management system 1 will be explained with reference to Figures 9 to 11 (and Figures 1 to 8 as appropriate). Figure 9 is an example flowchart showing the process in calibration mode. Figure 10 is an example flowchart showing the process in learning mode. Figure 11 is an example flowchart showing the process in particle size distribution estimation mode. (Operation related to calibration mode) Referring to Figure 9, the operation of the calibration mode will be explained. First, the operator flies, for example, the mobile unit 10 in intelligent flight mode, and the imaging unit 12 captures a calibration image of the spreading surface from above. The calibration image input unit 41 receives the captured calibration image as input (step S11). Next, the calibration parameter generation unit 42 creates calibration parameters based on the input calibration image (step S12). The created calibration parameters are stored in the storage unit 43 and used when operating in particle size distribution estimation mode. It is recommended not to change the intelligent flight mode settings or the environment related to the imaging unit 12's imaging after the calibration parameters have been created.

[0031] (Operation related to learning mode) Referring to Figure 10, the operation in learning mode will be explained. After completing the operation in calibration mode, the operator performs the operation in learning mode. First, the operator operates the shooting unit 12 to capture training images while the mobile unit 10 is flying in intelligent flight mode, and the training image input unit 51 starts inputting training images (step S21). The training image splitting unit 52 splits the training images and passes them to the training data generation unit 53 (step S22). The training data generation unit 53 displays the split training images 8a obtained from the training image splitting unit 52 on the screen, accepts the setting of tags (for example, information indicating that it is a component 9a) by the operator, and creates training data (step S23). The training data generation unit 53 creates multiple training data using, for example, captured images 8 from different areas. Next, the deep learning unit 54 trains a learning model (for example, MASK R-CNN) by deep learning using the training data created in step S23 (step S24). This creates a trained model (including training parameters), and the created trained model is stored in the memory unit 55 (step S25). The trained model detects components 9a (such as aggregates) contained in the material 9 by inputting the captured image 8 or its segmented image 8a of the material 9 captured by the imaging unit 12.

[0032] (Operation related to particle size distribution estimation mode) Referring to Figure 11, the operation of the particle size distribution estimation mode will be explained. After completing the operations related to the calibration mode and the learning mode, the operator performs the operation in the particle size distribution estimation mode. First, the operator operates the imaging unit 12 while the mobile unit 10 is flying in intelligent flight mode to capture an estimation image, and the estimation image input unit 61 starts inputting the estimation image (step S31). The estimation image splitting unit 62 splits the estimation image and passes it to the component detection unit 63 (step S32). The component detection unit 63 uses the trained model learned by deep learning in step S24 to detect components 9a (aggregates) from the split images 8a obtained by splitting the estimation image (step S33). In the detection in step S33, the split images 8a of the estimation image are input to the trained model (e.g., MASK R-CNN), and the detection result (information on the detected components 9a (aggregates, etc.)) output from the trained model is obtained. The recognition image merging unit 64 merges the segmented images 8a after detecting the components 9a (aggregates, etc.) to return them to the size of the original captured image 8 (step S34).

[0033] Next, the diameter determination unit 66 determines the diameter of the component 9a (aggregate) in the captured image 8 based on the component 9a (aggregate) detected in step S33 (step S35A). Alternatively, the diameter determination unit 66 may, for example, display the detection result on a display unit to allow the worker to select both ends of the component 9a, and then determine the actual diameter of the component 9a (aggregate) based on the number of pixels between the selected ends and calibration parameters. Furthermore, the area calculation unit 65 calculates the area of ​​each component 9a (aggregate) detected in step S33 (step S35B). Next, the mass conversion unit 67 calculates a mass conversion coefficient for each particle size category (step S36). Subsequently, the particle size distribution estimation unit 68 estimates the particle size distribution based on the total area of ​​each particle size category of the constituent material 9a (aggregate) and the mass conversion coefficient (step S37). Then, the particle size distribution estimation unit 68 displays the estimated particle size distribution on a display unit (not shown), and the particle size distribution confirmation alarm unit 69 determines whether the particle size distribution of each area falls within a predetermined control range, and if there is an area that does not fall within the predetermined control range, it notifies an alarm via the notification unit 23.

[0034] As described above, in the construction management system 1 according to the first embodiment, the material 9 after it has been spread is photographed area by area to estimate the particle size distribution. The particle size distribution estimated for each area can serve as an indicator of the bias in the placement of the material 9 in each area. Therefore, the bias of the constituent materials 9a can be quantitatively judged based on the particle size distribution estimated for each area. As a result, there is no room for human subjective judgment, and more accurate construction management is possible. Furthermore, in the construction management system 1 according to the first embodiment, the material 9 is photographed while moving along the spreading surface using the movable unit 10. Therefore, it is possible to easily manage construction over a wide area.

[0035] [Second Embodiment] <Configuration of the construction management system according to the second embodiment> Referring to Figure 12, the configuration of the construction management system 101 according to the second embodiment will be described. Figure 12 is a schematic diagram of the construction management system 101 according to the second embodiment. As shown in Figure 12, the construction management system 101 comprises an imaging unit 12 and a processing unit 20. The imaging unit 12 is mounted on a mobile unit 110 that can move on the spreading surface. In this embodiment, the mobile unit 110 is a construction machine, such as a bulldozer used in the spreading process or a vibratory roller used in the compaction process after the spreading process. The imaging unit 12 is installed on the front upper part of the mobile unit 110 and photographs the spreading surface in the direction of travel of the mobile unit 110. This allows the spreading surface to be photographed during the spreading or compaction process. The construction management system 101 according to the second embodiment described above can also achieve substantially the same effects as the first embodiment.

[0036] [Third Embodiment] <Configuration of the construction management system according to the third embodiment> Referring to Figure 13, the configuration of the construction management system 201 according to the third embodiment will be described. Figure 13 is a schematic diagram of the construction management system 201 according to the third embodiment. As shown in Figure 13, the construction management system 201 comprises an imaging unit 12 and a processing unit 20, with the imaging unit 12 being installed at a position higher than the surface from which the material is spread. The imaging unit 12 preferably has a mechanism that allows the imaging direction to be changed. The construction management system 201 according to the third embodiment described above can also achieve substantially the same effects as the first embodiment.

[0037] While embodiments of the present invention have been described so far, the present invention is not limited thereto and can be implemented without changing the spirit of the claims. For example, in each embodiment, MASK R-CNN was used as an example to explain the learning model (network) corresponding to instance segmentation. However, if it is possible to individually detect the target to be estimated from the captured image 8 on a pixel-by-pixel basis, other learning models can be used. [Explanation of symbols]

[0038] 1,101,201 Construction Management System 10,110 Mobile Unit 11 IMU 12 Photography Department 13 GPS 14 Communications Department 15 Control section 20 Processing Units 21 Communications Department 22 Analysis Department 23 Notification Department 30 Relay section 41 Calibration Image Input Section 42 Calibration parameter generation unit 43 Storage section 51 Learning Image Input Section 52. Image segmentation section for learning 53 Training Data Generation Unit 54 Deep Learning Department 55 Storage section 61 Image input section for estimation 62 Image division section for estimation 63 Component detection unit 64. Understanding the junction of portraits 65. Area Calculation Section 66. Path Determination Section 67. Quality Conversion Department 68. Particle size distribution estimation section 69 Particle size distribution confirmation alarm unit

Claims

1. A construction management system for managing the construction of the spreading process, The filming crew photographs the scattered materials area by area, The system includes a processing unit that analyzes captured images to obtain information regarding construction quality control for each area, The aforementioned processing unit, A component detection unit for detecting the components of the material in the captured image, A particle size distribution estimation unit that estimates the particle size distribution of the material from the detected components, It includes a particle size distribution determination unit that determines whether the particle size distribution of each area falls within a predetermined control range. A construction management system characterized by the following features.

2. The mobile unit on which the aforementioned imaging unit can be mounted is further provided. The aforementioned imaging unit moves along the dispensing surface and takes images of the material. The construction management system according to feature 1.

3. The aforementioned mobile unit is an unmanned aerial vehicle, which moves at a predetermined speed while maintaining a constant distance to the dispensing surface. The aforementioned imaging unit photographs the dispensing surface at predetermined time intervals. The construction management system according to claim 2, characterized by the features described above.

4. The moving part is a construction machine used in the spreading process or a construction machine used in the compaction process performed after the spreading process. The aforementioned imaging unit photographs the spreading surface during the spreading process or the compaction process. The construction management system according to claim 2, characterized by the features described above.

5. The component detection unit individually detects the components in the captured image using a trained model that has learned the components using a learning model corresponding to instance segmentation. The construction management system according to feature 1.

6. The aforementioned trained model has been machine-trained to detect the components by inputting segmented images obtained by dividing the captured image. The construction management system according to claim 5, characterized in that it is a construction management system.

7. A construction management method for managing the construction of the spreading process, The process involves photographing the scattered materials area by area, The system includes a processing step that analyzes captured images to obtain information regarding construction quality control for each area. The aforementioned processing step is: A component detection step for detecting the components of the material in the captured image, A particle size distribution estimation step for estimating the particle size distribution of the material from the detected components, The system includes a particle size distribution determination step that determines whether the particle size distribution of each area falls within a predetermined control range. A construction management method characterized by the following features.

Images