Underground pile foundation construction progress recognition method based on 2D image and video fusion

Through the fusion of 2D images taken by drone and video data, combined with deep learning algorithms, the construction progress of underground pile foundations is automatically identified and marked, which solves the problem of time-consuming and labor-intensive traditional manual statistics and achieves efficient construction progress tracking.

CN115909094BActive Publication Date: 2024-11-05HUAZHONG UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211319741.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-10-26
Publication Date
2024-11-05
Estimated Expiration
2042-10-26

AI Technical Summary

Technical Problem

Traditional manual statistics on the construction progress of underground pile foundations is time-consuming and labor-intensive, and the underground construction site is complex and hidden, making it difficult to achieve automated identification and tracking.

Method used

The drone is used to take 2D images and fuse them with the video data. The mechanical equipment and pile hole information is identified through image stitching, matching and deep learning algorithms, and the construction progress is automatically calculated and marked on the image.

Benefits of technology

Automatic identification and tracking of underground pile foundation construction progress is realized, construction efficiency is improved, accuracy is high and manual intervention is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115909094B_ABST
    Figure CN115909094B_ABST
Patent Text Reader

Abstract

The present invention belongs to the technical field related to the identification of the progress of underground pile foundation construction, and discloses a method for identifying the progress of underground pile foundation construction based on 2D image and video fusion, the method comprising: taking 2D images of the underground pile foundation construction site, identifying and marking the mechanical equipment information and construction pile hole information in the working area; performing image stitching on the taken 2D images to obtain a stitched image; dividing the working area on the stitched image, and marking the position and viewing angle information of the construction site camera; matching the mechanical equipment and construction pile hole in the construction site camera with those in the stitched image; obtaining the key frames of each mechanical equipment in the video taken by the construction site camera, and then obtaining the total working time of each mechanical equipment; obtaining the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile hole and marking it on the stitched image. The present application can realize the automatic identification and tracking of the progress of underground pile foundation construction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field related to underground pile foundation construction progress identification, and more specifically, relates to an underground pile foundation construction progress identification method based on 2D image and video fusion. Background Art

[0002] Image recognition refers to the technology of using computers to process, analyze and understand images in order to identify various different targets and objects. It is a practical application of deep learning algorithms. Among them, image classification recognition and image segmentation recognition can be used in engineering to ensure construction safety and improve construction efficiency.

[0003] However, underground installation construction sites are complex, located at a certain depth underground, and the construction is concealed. The traditional manual construction progress statistics method mainly relies on daily data for summary statistics, and marks the underground pile foundation construction progress in traditional drawings. However, this traditional manual method requires manual work to keep an eye on the work site, which is not only time-consuming but also laborious. Therefore, there is an urgent need to design a new underground pile foundation construction progress identification and tracking method. Summary of the invention

[0004] In view of the above defects or improvement needs of the prior art, the present invention provides a method for identifying the progress of underground pile foundation construction based on 2D image and video fusion, which can realize automatic identification and tracking of the progress of underground pile foundation construction.

[0005] To achieve the above-mentioned purpose, according to one aspect of the present invention, a method for identifying the progress of underground pile foundation construction based on 2D image and video fusion is provided, the method comprising: S1: using a drone to take 2D images of an underground pile foundation construction site, and identifying and marking the mechanical equipment information and construction pile hole information in the working area; S2: stitching the taken 2D images to obtain a stitched image, that is, a complete underground pile foundation construction site image; S3: dividing the working area of ​​the image underground pile foundation construction site on the stitched image, and marking the position and viewing angle information of the construction site camera in each divided area; S4: matching the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image; S5: obtaining the start running frame and stop running frame of each mechanical equipment in the video taken by the construction site camera, as well as the number and position information of the construction pile holes, and then obtaining the total working time of each mechanical equipment; S6: obtaining the progress of underground pile foundation construction based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and marking the construction progress on the stitched image.

[0006] Preferably, in step S5, the starting running frame and the stopping running frame are obtained in the following manner: a frame with the largest change in the mechanical equipment is selected as the starting frame of the duration record, i.e., the starting running frame, until there is no change for a long time, and the previous frame with no change is selected as the ending frame of the duration record, i.e., the stopping running frame.

[0007] Preferably, in step S5, the starting running frame and the stopping running frame are obtained in the following manner: S51: extracting mutual information features of video frames, and selecting two frames with the largest mutual information for k-value clustering; S52: calculating the corresponding pixel position change difference in two adjacent frame images, the pixel position change difference includes the pixel color difference, saturation difference and brightness difference of the two adjacent frames, and after obtaining the statistical histogram of each frame, taking the average value, and selecting the frame closest to the average value in the video frames after clustering in step S51 as the key frame of this category; S53: arranging the obtained key frames of various types in time sequence, calculating the Mahalanobis distance between the RGB mean matrices of every two frames to perform inter-frame similarity calculation, selecting a frame with a smaller inter-frame similarity with the previous frame and a larger inter-frame similarity with the next frame as the starting running frame, selecting a frame with a larger inter-frame similarity with the previous frame and a smaller inter-frame similarity with the next frame as the stopping running frame, and recording the duration between the two frames as the working duration.

[0008] Preferably, step S4 matches the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image. The specific steps are: S41: determine the location of the drone shooting based on the latitude, longitude and altitude information in the drone shooting image attributes, and obtain the location information of the target mechanical equipment based on the drone rotation angle and pitch angle; S42: determine the selected construction pile holes within the angle range of the construction site camera in combination with the construction site camera position and viewing angle information marked in the stitched image; S43: obtain the key frame of the equipment for preparing pile holes in the construction site camera, use the deep learning algorithm to perform depth estimation, and then determine the position of the equipment for preparing pile holes. The position of the equipment for preparing pile holes is the construction pile hole position, thereby achieving the matching of the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image.

[0009] Preferably, step S2 specifically includes the following steps: S21: normalizing and grayscale processing the 2D images taken by the drone in sequence; S22: automatically extracting feature points in the image based on the SIFT algorithm, and calculating the directions of the feature points; S23: performing feature point matching and coordinate transformation on multiple 2D images to achieve image stitching.

[0010] Preferably, step S22 specifically includes: S22a: performing extreme point detection on the 2D image in the Gaussian difference pyramid system space to obtain discrete extreme points at all scales; and using the curve fitting method to obtain a continuous space curve to locate the feature points; S22b: calculating the amplitude and direction of the gradient of the pixel in the 3σ neighborhood window of the feature point position, performing Gaussian coefficient weighted calculation on the corresponding amplitude, performing statistics on the direction, and obtaining a statistical histogram of the direction, and taking the direction of the histogram peak as the main direction of the feature point; S22c: rotating the neighborhood image according to the main direction, and then performing region division, the same as the principle of S22b, calculating the statistical histogram of each direction, and taking the direction of the histogram peak as the neighborhood direction of the feature point; S22d: normalizing the description vector of the feature point, and then setting a threshold for thresholding, sorting the description vector according to the scale of the feature point, and finally obtaining a series of feature points.

[0011] Preferably, identifying and labeling the mechanical equipment information and construction pile hole information in the working area in step S1 includes: using the trained GoogLeNet model to identify and label the 2D image taken by the drone.

[0012] Preferably, step S6 also includes progress comparison, which is used to compare the actual progress with the planned progress.

[0013] Preferably, the changes of mechanical equipment are identified based on the movements of the vertical state machine arm of the rotary drilling rig, the lifting arms of the truck crane and crawler crane, the rear of the concrete mixer truck and the casing.

[0014] According to another aspect of the present invention, a system for identifying the progress of underground pile foundation construction based on 2D image and video fusion is provided, the system comprising: an identification and annotation module: used to use a drone to take 2D images of the underground pile foundation construction site, and identify and annotate the mechanical equipment information and construction pile hole information in the working area; a splicing module: used to splice the captured 2D images to obtain a spliced ​​image, that is, a complete underground pile foundation construction site image; a division and annotation module: used to divide the working area of ​​the image underground pile foundation construction site on the spliced ​​image, and annotate the position and viewing angle information of the construction site camera in each divided area; a matching module: used to match the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the spliced ​​image; a calculation module: used to obtain the start running frame and the stop running frame of each mechanical equipment in the video taken by the construction site camera, as well as the number and position information of the construction pile holes, and then obtain the total working time of each mechanical equipment; a marking module: used to obtain the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and annotate the construction progress on the spliced ​​image.

[0015] In general, compared with the prior art, the above technical solution conceived by the present invention has the following beneficial effects in the underground pile foundation construction progress identification method based on 2D and video fusion:

[0016] 1. Based on the fusion of UAV aerial 2D images and video images, the mechanical equipment and pile hole information in the video are matched with those in the image. Then, the movement duration of the mechanical equipment can be determined by the key frame data in the video, and the construction progress can be determined by the movement duration, thereby realizing annotation in the 2D stitched image.

[0017] 2. The operation of the mechanical equipment is determined in the form of key frames, and the movement duration of the mechanical equipment is determined by the statistical histogram represented at the pixel level. The similarity is further calculated through the RGB mean matrix to obtain the start running frame and the stop running frame in the key frame, and then the running duration is obtained according to the difference between the start running frame and the stop running frame, with high accuracy and complete statistics.

[0018] 3. By identifying the key frames of the equipment in the video and determining the pile hole position based on the deep learning algorithm, the mechanical equipment and construction pile holes in the construction site camera are matched with the mechanical equipment and construction pile holes in the stitched image based on the position and viewing angle information of the construction site camera used for video shooting. The matching of 2D images and video data is achieved, making it possible for subsequent annotation in 2D images through video data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the underground pile foundation construction progress identification method based on 2D image and video fusion in this embodiment. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0021] The present invention provides a method for identifying the progress of underground pile foundation construction based on 2D image and video fusion. The method comprises the following steps S1 to S6.

[0022] S1: Use drones to capture 2D images of the underground pile foundation construction site, and identify and mark the mechanical equipment information and construction pile hole information in the work area.

[0023] During the drone aerial photography process, the drone is first installed and calibrated as required to ensure flight safety and position accuracy. Then, the aerial surround mode is selected to capture images of the construction site. Furthermore, the pitch angle of the camera on the drone is adjusted. Finally, the heading overlap rate and lateral overlap rate of the drone are set to ensure the accuracy of the construction of the underground pile foundation construction scene model.

[0024] In this embodiment, in order to ensure that the recognition of 2D images in the drone images can be realized, at least the drone 45° aerial images are included. The mechanical equipment used in the underground pile foundation construction process mainly includes rotary drilling rigs, excavators, truck cranes, crawler cranes, concrete mixers, etc. Among them, the most important is the rotary drilling rig, which is used to prepare the construction pile hole.

[0025] Before using the images taken by drones for mechanical equipment recognition, the image size needs to be unified, the image needs to be enhanced, and the image brightness, contrast, saturation, etc. need to be adjusted. If necessary, the image rotation angle, horizontal displacement, vertical displacement, etc. also need to be adjusted. If the GoogLeNet model has not been trained, it is necessary to train the GoogLeNet model with part of the data set, update the model network parameters according to the loss value obtained from the neural network training, fine-tune the network parameters and save the network parameters, and perform iterative training until the loss function converges. Based on the GoogLeNet model, mechanical equipment such as rotary drilling rigs, excavators, truck cranes, crawler cranes, and concrete mixers are annotated, and the location information of the mechanical equipment in the image is marked.

[0026] S2: stitching the captured 2D images to obtain a stitched image, that is, a complete underground pile foundation construction site image.

[0027] Step S2 specifically includes the following steps:

[0028] S21: The 2D images taken by the drone are normalized and grayscale processed in sequence.

[0029] The image is normalized to avoid uneven brightness in the image due to different shooting angles of drone aerial photography.

[0030] Grayscale processing is performed because there is a lot of soil attached to the surface of mechanical equipment and the boundaries of the work area in the drone aerial photos of the underground pile foundation construction site, which is easy to blend into the background. Therefore, grayscale processing is required to avoid insufficient image contrast and thus affect subsequent image recognition.

[0031] S22: Automatically extract feature points in the image based on the SIFT algorithm and calculate the directions of the feature points.

[0032] The specific steps are:

[0033] S22a: Detect extreme points in the Gaussian difference pyramid system space to obtain discrete extreme points at all scales; use curve fitting method to obtain continuous space curves to locate feature points; filter feature points to eliminate the influence of edge points;

[0034] S22b: Calculate the main direction of the feature point: Calculate the amplitude and direction of the gradient of the pixel in the 3σ neighborhood window of the feature point position, perform Gaussian coefficient weighted calculation on the corresponding amplitude, perform statistics on the direction, and obtain a statistical histogram of the direction, and take the direction of the histogram peak as the main direction of the feature point;

[0035] S22c: Calculate the neighborhood direction of the feature point: rotate the neighborhood image according to the main direction, and then divide the area. The principle is the same as S22b. Calculate the statistical histogram of each direction, and use the direction of the histogram peak as the neighborhood direction of the feature point;

[0036] S22d: Normalize the description vector of the feature point, then set a threshold to perform threshold processing, sort the description vector according to the scale of the feature point, and finally obtain a series of feature points.

[0037] S23: Perform feature point matching and coordinate transformation on multiple 2D images to achieve image stitching.

[0038] S3: Divide the working area of ​​the underground pile foundation construction site on the stitched image, and mark the position and viewing angle information of the construction site camera in each divided area.

[0039] According to the actual situation of the construction site, the working area of ​​the underground pile foundation construction site is divided in the stitched image, and the position and viewing angle information of the construction site camera in each divided area are marked.

[0040] S4: Match the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image.

[0041] The specific steps are:

[0042] S41: determining the location of the drone shooting according to the latitude, longitude and altitude information in the attributes of the drone shot image, and obtaining the location information of the target mechanical equipment according to the rotation angle and pitch angle of the drone;

[0043] S42: Determine a candidate construction pile hole within the angle range of the construction site camera by combining the construction site camera position and viewing angle information marked in the stitched image;

[0044] S43: Acquire the key frame of the equipment for preparing pile holes in the construction site camera, use the deep learning algorithm to perform depth estimation, and then determine the position of the equipment for preparing pile holes. The position of the equipment for preparing pile holes is the position of the construction pile hole, so as to achieve matching between the mechanical equipment and the construction pile hole in the construction site camera and the mechanical equipment and the construction pile hole in the stitched image.

[0045] S5: Obtain the start running frame and stop running frame of each mechanical equipment in the video captured by the construction site camera, as well as the number and position information of the construction pile holes, and then obtain the total working time of each mechanical equipment.

[0046] The vertical arm of a rotary drilling rig, the lifting arms of a truck crane and crawler crane, the rear of a concrete mixer truck and the casing are taken as research objects to identify the changes of mechanical equipment in the video.

[0047] The start running frame and stop running frame can be obtained as follows:

[0048] The frame with the largest change in the mechanical equipment is selected as the starting frame of the duration recording, that is, the starting running frame, until there is no change for a long time, and the previous frame without change is selected as the ending frame of the duration recording, that is, the stopping running frame.

[0049] You can also use the following method to get the start running frame and stop running frame:

[0050] S51: Perform mutual information feature extraction on the video frames, and select the two frames with the largest mutual information for k-value clustering. Perform k-value clustering on the video frame sequence, and select a custom k value, which can limit the number of key frames to a certain extent, effectively reduce redundancy, and represent the main content of the shot. Mutual information feature extraction is often used to measure and count the correlation between two random variables, and select the two frames with the largest mutual information for clustering.

[0051] S52: Calculate the position change difference of corresponding pixels in two adjacent frames of images, where the pixel position change difference includes the color difference, saturation difference and brightness difference of pixels in two adjacent frames, obtain the statistical histogram of each frame, take the average value, and select the frame closest to the average value among the video frames clustered in step S51 as the key frame of this class;

[0052] S53: Arrange the obtained key frames in time sequence, calculate the Mahalanobis distance between the RGB mean matrices of every two frames, calculate the inter-frame similarity, select a frame with a smaller inter-frame similarity with the previous frame and a larger inter-frame similarity with the next frame as the start running frame, select a frame with a larger inter-frame similarity with the previous frame and a smaller inter-frame similarity with the next frame as the stop running frame, and record the duration between the two frames as the working duration, including but not limited to the drilling duration of the rotary drilling rig, the soil clearing duration of the excavator, the hoisting duration of the truck crane and crawler crane, and the concrete pouring duration of the concrete mixer truck.

[0053] S6: Obtaining the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and marking the construction progress on the spliced ​​image.

[0054] The carrying capacity or workload per unit time of each mechanical equipment is known, so the construction progress of the underground pile foundation can be obtained according to the working time of the mechanical equipment and the number and location information of the construction pile holes. The construction progress is marked on the spliced ​​image to obtain the marked spliced ​​2D image.

[0055] Progress analysis can also be performed to compare the actual progress with the planned progress. The planned progress is divided according to project nodes and construction site areas. Each construction node in each specific construction area corresponds to a sub-plan progress. The sub-plan progress includes the type and quantity of construction equipment in the work area, the working hours and working status of the construction equipment, etc. When the actual progress is compared with the planned progress, if the actual progress is the same as the planned progress, the actual progress meets the planned requirements. If the actual progress is different from the planned progress, the actual progress does not meet the planned requirements, and the construction process should be accelerated or slowed down accordingly.

[0056] On the other hand, the present application provides an underground pile foundation construction progress recognition system based on 2D image and video fusion, the system comprising:

[0057] Identification and labeling module: used to use drones to take 2D images of underground pile foundation construction sites, and identify and label the mechanical equipment information and construction pile hole information in the work area;

[0058] Stitching module: used to stitch the captured 2D images to obtain a stitched image, that is, a complete underground pile foundation construction site image;

[0059] Division and annotation module: used to divide the working area of ​​the underground pile foundation construction site on the stitched image, and to annotate the location and viewing angle information of the construction site camera in each divided area;

[0060] Matching module: used to match the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image;

[0061] Calculation module: used to obtain the start running frame and stop running frame of each mechanical equipment in the video shot by the construction site camera, as well as the number and location information of the construction pile holes, and then obtain the total working time of each mechanical equipment;

[0062] Annotation module: used to obtain the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and to annotate the construction progress on the spliced ​​image.

[0063] The system also includes a progress analysis module for implementing the above-mentioned progress analysis process. In the progress analysis, if the actual progress of the work area meets the planned requirements, the background is light green. If the actual progress of the work area does not meet the planned requirements, the background is light red. At the same time, human-computer interaction can be carried out to obtain information on each construction area, visualize the progress of each construction area of ​​the current underground pile foundation construction, and realize automatic identification and tracking of progress.

[0064] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying the progress of underground pile foundation construction based on 2D image and video fusion, characterized in that: The method comprises: S1: Use drones to take 2D images of the underground pile foundation construction site, identify and mark the mechanical equipment information and construction pile hole information in the work area; S2: stitching the captured 2D images to obtain a stitched image, i.e., a complete image of the underground pile foundation construction site; S3: Divide the working area of ​​the underground pile foundation construction site on the stitched image, and mark the position and viewing angle information of the construction site camera in each divided area; S4: matching the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image; S5: Obtain the start running frame and stop running frame of each mechanical equipment in the video captured by the construction site camera, as well as the number and position information of the construction pile holes, and then obtain the total working time of each mechanical equipment; S6: Obtaining the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and marking the construction progress on the spliced ​​image; Step S4 matches the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the spliced ​​image. The specific steps are: S41: determining the location of the drone shooting according to the latitude, longitude and altitude information in the attributes of the drone shot image, and obtaining the location information of the target mechanical equipment according to the rotation angle and pitch angle of the drone; S42: Determine a candidate construction pile hole within the angle range of the construction site camera by combining the construction site camera position and viewing angle information marked in the stitched image; S43: Acquire the key frame of the equipment for preparing pile holes in the construction site camera, use the deep learning algorithm to perform depth estimation, and then determine the position of the equipment for preparing pile holes. The position of the equipment for preparing pile holes is the position of the construction pile hole, so as to achieve matching between the mechanical equipment and the construction pile hole in the construction site camera and the mechanical equipment and the construction pile hole in the stitched image.

2. The method according to claim 1, characterized in that In step S5, the start running frame and the stop running frame are obtained in the following manner: The frame with the largest change in the mechanical equipment is selected as the starting frame of the duration recording, that is, the starting running frame, until there is no change for a long time, and the previous frame without change is selected as the ending frame of the duration recording, that is, the stopping running frame.

3. The method according to claim 1, characterized in that In step S5, the start running frame and the stop running frame are obtained in the following manner: S51: extract mutual information features from the video frames, and select two frames with the largest mutual information for k-value clustering; S52: Calculate the position change difference of corresponding pixels in two adjacent frames of images, where the pixel position change difference includes the color difference, saturation difference and brightness difference of pixels in two adjacent frames, obtain the statistical histogram of each frame, take the average value, and select the frame closest to the average value among the video frames clustered in step S51 as the key frame of this class; S53: Arrange the obtained key frames in time sequence, calculate the Mahalanobis distance between the RGB mean matrices of every two frames to calculate the similarity between frames, select a frame with a smaller similarity to the previous frame and a larger similarity to the next frame as the start running frame, select a frame with a larger similarity to the previous frame and a smaller similarity to the next frame as the stop running frame, and record the duration between the two frames as the working duration.

4. The method according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21: normalizing and grayscale processing the 2D images taken by the drone in sequence; S22: Automatically extract feature points in the image based on the SIFT algorithm and calculate the direction of the feature points; S23: Perform feature point matching and coordinate transformation on multiple 2D images to achieve image stitching.

5. The method according to claim 4, characterized in that Step S22 specifically includes: S22a: Detect extreme points of the 2D image in the Gaussian difference pyramid scale space to obtain discrete extreme points at all scales; and use the curve fitting method to obtain a continuous space curve to locate feature points; S22b: Calculate the amplitude and direction of the gradient of the pixel in the 3σ neighborhood window of the feature point position, perform Gaussian coefficient weighted calculation on the corresponding amplitude, perform statistics on the direction, obtain a statistical histogram of the direction, and take the direction of the histogram peak as the main direction of the feature point; S22c: Rotate the neighborhood image in the main direction and then divide the area. The principle is the same as S22b. Calculate the statistical histogram of each direction and take the direction of the histogram peak as the neighborhood direction of the feature point. S22d: Normalize the description vector of the feature point, then set a threshold to perform threshold processing, sort the description vector according to the scale of the feature point, and finally obtain a series of feature points.

6. The method according to claim 1, characterized in that In step S1, identifying and labeling the mechanical equipment information and the construction pile hole information in the working area includes: using the trained GoogLeNet model to identify and label the 2D image taken by the drone.

7. The method according to claim 1, characterized in that Step S6 also includes progress comparison, which is used to compare the actual progress with the planned progress.

8. The method according to claim 2 or 3, characterized in that: Identify changes in mechanical equipment based on the movements of the vertical arm of the rotary drilling rig, the crane arms of the truck crane and crawler crane, the rear of the concrete mixer truck and the casing.

9. An underground pile foundation construction progress recognition system based on 2D image and video fusion, characterized in that: The system comprises: Identification and labeling module: used to use drones to take 2D images of underground pile foundation construction sites, and identify and label the mechanical equipment information and construction pile hole information in the work area; Stitching module: used to stitch the captured 2D images to obtain a stitched image, that is, a complete underground pile foundation construction site image; Division and annotation module: used to divide the working area of ​​the underground pile foundation construction site on the stitched image, and to annotate the location and viewing angle information of the construction site camera in each divided area; Matching module: used to match the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image; Calculation module: used to obtain the start running frame and stop running frame of each mechanical equipment in the video shot by the construction site camera, as well as the number and location information of the construction pile holes, and then obtain the total working time of each mechanical equipment; Annotation module: used to obtain the underground pile foundation construction progress based on the working time of each mechanical equipment and the number and position information of the construction pile holes, and annotate the construction progress on the spliced ​​image; The matching module matches the mechanical equipment and construction pile holes in the construction site camera with the mechanical equipment and construction pile holes in the stitched image. The specific steps are as follows: Determine the location of the drone shooting based on the latitude, longitude and altitude information in the drone image attributes, and obtain the location information of the target mechanical equipment based on the drone's rotation angle and pitch angle; Determine the candidate construction pile holes within the angle range of the construction site camera by combining the construction site camera position and viewing angle information marked in the stitched image; The key frames of the equipment for preparing pile holes in the construction site camera are obtained, and the depth is estimated using a deep learning algorithm to determine the position of the equipment for preparing pile holes. The position of the equipment for preparing pile holes is the position of the construction pile holes, thereby achieving matching between the mechanical equipment and the construction pile holes in the construction site camera and those in the stitched image.