A construction site work center estimation method and system

Abstract

The application discloses a construction site work center estimation method and system, the method comprises the following steps: S01, obtaining the image of the construction site through the image acquisition unit, and calibrating the image acquisition unit to obtain the calibration parameter; S02, detecting the key target in the image of the construction site, the key target comprises the head of the personnel, the safety helmet, the electric pole and the safety fence; S03, calculating the image barycenter of each key target in the current visual field to obtain the image barycenter set; S04, traversing the image barycenter set, obtaining the normalized coordinate of the image barycenter of each key target according to the calibration parameter, calculating the direction vector of the normalized coordinate under the image acquisition unit coordinate system, and then converting the direction vector into the unit polar coordinate vector; S05, clustering the unit polar coordinate vector to obtain the clustering center, which is the construction site work center. The application has the advantages of low cost, simple operation and the like.

Description

Technical Field

[0001] The present invention mainly relates to the field of visual technology, and in particular to a method and system for estimating a work center on a construction site. Background Art

[0002] The work center estimation method based on control ball scheduling uses control balls to detect key objects on the construction site, such as people, hard hats, safety fences, and utility poles. By integrating this information with the locations of these key objects, the high-risk field of view at the construction site is inferred as the work center. Estimating the construction site center has become an essential step in identifying supervisors who leave their posts, when work team members are out of their sight, identifying workers, and monitoring on-site safety measures.

[0003] Methods for estimating work centers can be divided into two categories: one is achieved through manual on-site electronic fence drawing, that is, manually marking an area on the video to represent the work area. However, electronic fences require manual determination of the work scene, and there is currently no clear demarcation standard. Different personnel have different judgments, which makes implementation difficult and the effectiveness of electronic fences low. The other method uses sensors on wearable devices such as hard hats, wristbands, and UWB ID cards, combined with small UWB base stations installed on-site, to determine the work scene by determining the relative positions of workers, safety tools, and security cameras through sensor signals. However, the combination of wearable devices and UWB base stations requires professional personnel to deploy a large amount of equipment on-site and debug the network, which is costly. Due to the large number of actual work sites, it has poor practicality and is difficult to widely apply. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: in response to the problems existing in the prior art, the present invention provides a construction site work center estimation method and system with low cost and easy operation.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] A construction site work center estimation method comprises the following steps:

[0007] S01. Acquire an image of the construction site through an image acquisition unit, and calibrate the image acquisition unit to obtain calibration parameters;

[0008] S02. Detect key targets in the image of the construction site, wherein the key targets include a person's head, a hard hat, a utility pole, and a safety fence;

[0009] S03, calculating the image center of gravity of each key target in the current field of view to obtain an image center of gravity set;

[0010] S04, traverse the image barycenter set, and get normalized coordinates by de-distorting the image barycenter of each key target according to the calibration parameters, calculate the direction vector of the normalized coordinates under the image acquisition unit coordinate system, and then convert the direction vector into a unit polar coordinate vector;

[0011] S05, cluster the unit polar coordinate vector to obtain a cluster center, which is the working center of the construction site.

[0012] Preferably, the specific process of obtaining normalized coordinates by de-distorting the image barycenter of each key target according to the calibration parameters in step S04 is as follows:

[0013] Suppose P i is the image barycenter of the key target, c x , c y is the image center, f x , f y are the focal lengths in two directions, distortion coefficients k1, k2, k3; the coordinate relationship between the points before and after distortion is as follows:

[0014] x1, y1 are normalized coordinates before distortion, and u, v are normalized coordinates after distortion:

[0015] x1=u(1+k1r 2 +k2r 4 +k3r 6 )

[0016] y1=v(1+k1r 2 +k2r 4 +k3r 6 )

[0017] r 2 =u 2 +v 2 , r is an unknown quantity, and the fixed point theorem is used to iteratively solve the normalized coordinates u and v.

[0018] Preferably, the specific process of using the fixed point theorem to iteratively solve the normalized coordinates u and v is as follows:

[0019] S41, normalize the pixel coordinates of the point before distortion:

[0020] x1=(x-c x ) / f x

[0021] y1=(y-c y ) / f y

[0022] S42, set the initial value of u and v as x1 and y1;

[0023] S43. Calculate currs=(1+k1r according to the values ​​of u and v. 2 +k2r 4 +k3r 6 )

[0024] S44, according to Update the values ​​of u and v;

[0025] S45. Repeat steps S43 and S44 multiple times to obtain the normalized coordinates u and v of the dedistorted point.

[0026] Preferably, in step S04, the specific process of calculating the direction vector whose normalized coordinates are located in the image acquisition unit coordinate system and then converting the direction vector into a unit polar coordinate vector is as follows:

[0027] Assume that the acquisition unit's X-direction scaling factor is k, the Y-direction scaling factor is l, and the focal length is f; the direction vector v in the image acquisition unit coordinate system is solved based on the dedistorted normalized coordinates, which has the following transformation relationship:

[0028] v*f=((u-cx) / k,(v-cy) / l,f)

[0029] f x =f*k

[0030] f y =f*l

[0031] Then: v=((u-cx) / fx,(v-cy) / fy,1)

[0032] Where fx and fy are the focal lengths in the x and y directions in pixel units; v is the direction vector in the camera coordinate system; converted to polar coordinates PT i There are the following transformation relationships:

[0033] p=arctan(v z / v x )

[0034] t=arccos(v y )

[0035] PT i =(p,t)

[0036] where v x ,v y ,v z is the value of the direction vector v in the x, y, and z directions; PT i is the polar coordinate unit vector in the camera coordinate system.

[0037] Preferably, the specific process of step S05 is:

[0038] S51, different clustering weights are given to different targets according to prior information;

[0039] S52, the weighted polar coordinate unit vectors are clustered by using a mean shift method to obtain clustering centers, wherein a kernel function used by the mean shift method is a Gaussian function.

[0040] Preferably, the image acquisition unit is a cloud camera of a surveillance ball in the construction site.

[0041] Preferably, in step S02, the safety fence in the image is segmented by using a deep learning segmentation network to realize detection of the safety fence.

[0042] The application further discloses a construction site work center estimation system, comprising:

[0043] A first program module is used for acquiring an image of a construction site by using an image acquisition unit, calibrating the image acquisition unit, and obtaining calibration parameters.

[0044] A second program module is used for detecting key targets in the image of the construction site, wherein the key targets include a head of a person, a safety helmet, a power pole and a safety fence.

[0045] A third program module is used for calculating image barycenters of the key targets in a current field of view to obtain an image barycenter set.

[0046] A fourth program module is used for traversing the image barycenter set, obtaining normalized coordinates of the image barycenters of each key target by removing distortion according to the calibration parameters, calculating direction vectors of the normalized coordinates in a coordinate system of the image acquisition unit, and converting the direction vectors into unit polar coordinate vectors.

[0047] A fifth program module is used for clustering the unit polar coordinate vectors to obtain clustering centers, which are work centers of the construction site.

[0048] The application further discloses a computer readable storage medium, which stores a computer program, wherein the computer program is used for executing steps of the method when being run by a processor.

[0049] The application further discloses a computer device, which comprises a memory and a processor, and the memory stores a computer program, wherein the computer program is used for executing steps of the method when being run by the processor.

[0050] Compared with the prior art, the application has the following advantages:

[0051] This method combines construction site control ball scheduling with analysis of target polar coordinate vector directional relationships to estimate the work center area. Specifically, the research objects of the control ball scheduling work center are hard hats, utility poles, and fences. For power construction sites, fences must be installed before construction begins, and workers entering the construction site must wear hard hats. The control ball is now dispatched to search the site, using the control ball to capture the on-site video stream, detect hard hats, worker heads, utility poles in the construction site video stream, and segment targets such as safety fences. By integrating key target location information, high-risk vision areas of the construction site are inferred, namely the work center. Areas with a large number of workers, safety fences, hard hats, and utility poles on the construction site are likely to be the work center area. This method can solve the problems of high cost, poor usability, and low practicality of existing technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 The present invention is a flowchart of an embodiment of the method.

[0053] Figure 2 This is a safety helmet, utility pole detection, and fence segmentation map in an embodiment of the present invention.

[0054] Figure 3 This is a working center effect diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0056] like Figure 1 As shown, the construction site work center estimation method according to an embodiment of the present invention includes the following steps:

[0057] S01. Acquire an image of the construction site through an image acquisition unit (such as a control ball at the construction site), and calibrate the image acquisition unit to obtain calibration parameters;

[0058] S02. Detect key targets in the image of the construction site, wherein the key targets include a person's head, a hard hat, a utility pole, and a safety fence;

[0059] S03, calculating the image center of gravity of each key target in the current field of view to obtain an image center of gravity set;

[0060] S04, traversing the image centroid set, dedistorting the image centroid of each key target according to the calibration parameters to obtain normalized coordinates, calculating the direction vector of the normalized coordinates in the image acquisition unit coordinate system, and then converting the direction vector into a unit polar coordinate vector;

[0061] S05. Cluster the unit polar coordinate vectors to obtain the cluster center, which is the construction site work center.

[0062] At the construction site, if there are many on-site workers, safety fences, hard hats, and electric poles, it is most likely the work center area. Based on the above facts, the present invention combines the construction site control ball scheduling, analyzes the target polar coordinate vector direction relationship, and estimates the work center area. Among them, the research objects of the control ball scheduling work center are safety helmets, electric poles, and fences; for power construction sites, fences need to be installed before construction begins, and workers need to wear safety helmets when entering the construction site; the control ball is now dispatched to search the site, and the control ball is used to obtain the on-site video stream, detect safety helmets, worker heads, electric poles in the construction site video stream, and segment targets such as safety fences; integrate key target location information to infer the high-risk field of view of the construction site, which is the work gathering center; if there are many on-site workers, safety fences, safety helmets, and electric poles, it is most likely the work center area. This method can solve the problems of high cost, poor usability, and low practicality in the existing technology.

[0063] In one specific embodiment, in step S01, the image acquisition unit is a pan-tilt camera or other imaging device used in a construction site surveillance system. The camera is calibrated using a checkerboard calibration method. This involves printing a checkerboard pattern on the ground and revising the camera calibration parameters, including intrinsic and extrinsic parameters, and distortion coefficients, based on the relationships between the checkerboard patterns.

[0064] In a specific embodiment, in step S02, the detection of a person's head, a helmet, and a utility pole is not limited to a deep learning detection network, and other detection networks can also be used; the detection of a safety fence is not limited to a deep learning segmentation network, and other segmentation networks can also be used. The steps for locating the fence area at the construction site are:

[0065] 1) Extract the red fence portion from the construction site image based on the fence color feature (e.g., the fence is red and white), then convert it into a binary image to train the ENet network, obtaining a trained ENet network model;

[0066] 2) Use the trained ENet network model to detect whether there is a fence in the construction site image. If there is no fence, obtain the next construction site image. If there is a fence, segment the fence in the construction site image to obtain the red fence in the fence;

[0067] 3) Based on the segmented red fence, the foreground pixels containing the red fence are filled with rectangles and expanded with polygons to generate a surface area covering the entire fence. At the same time, the positions corresponding to the surface areas in the construction site image are marked to achieve the positioning of the construction site fence area.

[0068] The above method uses fences with distinct color features to train a neural network. The segmentation results are then processed using a computer graphics algorithm. This method can quickly and accurately segment the location of construction site fences, offering the advantages of low cost, high accuracy, and low computational complexity. By algorithmically processing the partially segmented red fences, the entire fence area is determined and the fences are identified within the construction site. The red and white color of the fences is incorporated into the data processing phase, bypassing traditional segmentation and data annotation methods. This training set is generated based on color features, significantly reducing labor and time resources compared to traditional methods. By using polygon expansion to connect the gaps between fences to form a single fence surface, the entire fence area is fitted from the segmented results. This method achieves construction site fence area localization from part to whole, from patch to surface, without consuming significant storage space or computational time. This reduces the time and cost of training with large, fully labeled training sets while accurately and quickly localizing the fence area, thereby ensuring the safety of personnel at power construction sites.

[0069] In a specific embodiment, in step S03, the calculation formula for calculating the center of gravity of the image of the person's head, safety helmet, telephone pole, and safety fence in the current field of view is:

[0070]

[0071]

[0072] where X c ,Y c is the image centroid of the target, and the image centroid of each target is recorded as set P.

[0073] In a specific embodiment, the specific implementation process of step S05 is:

[0074] The center of gravity P of the target image is known i , c x 、c y is the image center, f x , f y are the focal lengths in two directions, and the distortion coefficients k1, k2, and k3 respectively; the coordinate relationship of the points before and after the distortion is as follows:

[0075] x1, y1 are the normalized coordinates before distortion, u, v are the normalized coordinates after distortion

[0076] x1=u(1+k1r 2 +k2r 4 +k3r 6 )

[0077] y1=v(1+k1r2 +k2r 4 +k3r 6 )

[0078] Because r 2 =u 2 +v 2 , r is also an unknown quantity, and the fixed point theorem is used to iteratively solve u and v, specifically:

[0079] S41. First, normalize the pixel coordinates of the point before distortion:

[0080] x1=(xc x ) / f x

[0081] y1=(yc y ) / f y

[0082] S42. Given the initial values ​​of u and v as x1 and y1;

[0083] S43. Calculate currs=(1+k1r according to the values ​​of u and v. 2 +k2r 4 +k3r 6 );

[0084] S44, according to Update the values ​​of u and v

[0085] S45. Repeat steps S43-S44 five times to obtain the normalized coordinates u and v of the dedistorted point.

[0086] Given a camera with a known X-axis scaling factor of k, a Y-axis scaling factor of l, and a focal length of f, the direction vector v in the camera coordinate system can be obtained from the dedistorted coordinates using the following transformation relationship:

[0087] v*f=((u-cx) / k,(v-cy) / l,f)

[0088] f x =f*k

[0089] f y =f*l

[0090] Then v=((u-cx) / fx,(v-cy) / fy,1)

[0091] Where fx and fy are the focal lengths in the x and y directions in pixel units. v is the direction vector in the camera coordinate system. Convert to polar coordinates PT i There are the following transformation relationships:

[0092] p=arctan(v z / vx )

[0093] t=arccos(v y )

[0094] PT i =(p,t)

[0095] where v x ,v y ,v z Is the value of the direction vector v in the x, y, and z directions. PT i is the unit vector of polar coordinates in the camera coordinate system.

[0096] In a specific embodiment, the specific process of step S05 is:

[0097] S51, assigning different clustering weights to different targets based on prior information;

[0098] S52. Cluster the weighted polar coordinate unit vectors using a mean shift method to obtain cluster centers; the kernel function used in the mean shift method is a Gaussian function; and the formula of the mean shift method is:

[0099]

[0100] Where h is the bandwidth, x is the current cluster center, and x i is the polar coordinate unit quantity, c k,d / nh d is the unit density. Using this formula for clustering, we get the cluster center, which is the work center.

[0101] The method of the present invention is further described in detail below with reference to the accompanying drawings:

[0102] like Figure 1 As shown, the work center estimation method based on control ball scheduling of the present invention obtains construction site photos, detects safety helmets, personnel heads, telephone poles, and segmented safety fences in the target area in the construction site photos, calculates the direction vector of the target in the camera coordinate system, and converts it into a polar coordinate PT coordinate system, that is, a unit polar vector; the unit polar vector is weighted according to prior information, and then an unsupervised clustering method is used to cluster the unit polar vectors to obtain a cluster center, which is the estimated work center.

[0103] Based on the above ideas, we can get Figure 2As a result, the objects recognized by the boxes are telephone poles, workers, and fences; many targets of various shapes at the construction site are detected; the improved Yolo network is used to increase its generalization performance for specific targets and make it suitable for running on the control ball device to obtain the two-dimensional image coordinates of safety helmets, personnel heads, and telephone poles; the safety fences at the construction site are segmented and the improved ENet network is used to enable it to be quickly segmented on edge devices such as the control ball, reducing the computational complexity and obtaining the safety fence contour coordinates on the image.

[0104] Figure 3 The white area represents the work site found after the dispatch control ball is deployed. The points on the plan represent utility poles, workers, and fences, and the boxes represent the work center. The two-dimensional polar coordinates of the workers and utility poles can be obtained. The control ball is calibrated to obtain the camera intrinsic parameters, which are used to dedistort the detection and segmentation results. The target image position is calculated in the camera coordinate system using the camera imaging principle in combination with the camera intrinsic parameters and converted into a polar coordinate unit vector PT.

[0105] like Figure 3 As shown in the box, different cluster weights are assigned to different targets based on prior information. The weighted polar unit vectors are clustered using the mean-shift method, which uses a Gaussian kernel function. Clustering is performed using polar unit vector information to determine cluster centers, which are then inspected using a control ball.

[0106] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for those skilled in the art, various improvements and modifications that do not depart from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A construction site work center estimation method, characterized in that: Including steps: S01. Acquire an image of the construction site through an image acquisition unit, and calibrate the image acquisition unit to obtain calibration parameters; wherein the image acquisition unit is a pan-tilt camera of a control ball at the construction site; S02. Detect key targets in the image of the construction site, wherein the key targets include a person's head, a hard hat, a utility pole, and a safety fence; S03, calculating the image center of gravity of each key target in the current field of view to obtain an image center of gravity set; S04, traversing the image centroid set, dedistorting the image centroid of each key target according to the calibration parameters to obtain normalized coordinates, calculating the direction vector of the normalized coordinates in the image acquisition unit coordinate system, and then converting the direction vector into a unit polar coordinate vector; S05. Cluster the unit polar coordinate vectors to obtain the cluster center, which is the construction site work center; The steps for locating the safety fence area at the construction site are: 1) Extract the red fence portion of the construction site image based on the fence color features, then convert it into a binary image to train the ENet network to obtain a trained ENet network model; 2) Use the trained ENet network model to detect whether there is a fence in the construction site image. If there is a fence, segment the fence in the construction site image to obtain the red fence; 3) Based on the segmented red fence, the foreground pixels containing the red fence are filled with rectangles and expanded with polygons to generate a surface area covering the entire fence. The corresponding locations of the surface area in the construction site image are marked to locate the construction site fence area. In step S02, the safety fence in the image is segmented by a deep learning segmentation network to detect the safety fence.

2. The construction site work center estimation method according to claim 1, characterized in that: The specific process of dedistorting the image center of gravity of each key target to obtain normalized coordinates according to the calibration parameters in step S04 is as follows: Assumptions is the image center of gravity of the key target, is the center of the image, The focal length in two directions and the distortion coefficient are ; The coordinate relationship of the points before and after distortion is as follows: , is the normalized coordinate before distortion, , is the normalized coordinate after distortion: , r is the unknown quantity, and the fixed point theorem is used to iteratively solve the normalized coordinates u, v.

3. The construction site work center estimation method according to claim 2, characterized in that: The specific process of iteratively solving the normalized coordinates u and v using the fixed point theorem is: S41. Normalize the pixel coordinates of the point before distortion: S42, given u, the initial value of v is , ; S43, calculate according to the value of u, v S44, according to Update the values ​​of u and v; S45. Repeat steps S43 and S44 multiple times to obtain the normalized coordinates u and v of the dedistorted point.

4. The construction site work center estimation method according to claim 2 or 3, characterized in that: In step S04, the specific process of calculating the direction vector whose normalized coordinates are located in the image acquisition unit coordinate system and then converting the direction vector into a unit polar coordinate vector is as follows: Assume that the unit X-direction scaling factor is k and the Y-direction scaling factor is , the focal length is f; according to the normalized coordinates after dedistortion, the direction vector v in the image acquisition unit coordinate system has the following transformation relationship: but: in 、 is the focal length in the x and y directions in pixel units; v is the direction vector in the camera coordinate system; converted to polar coordinates There are the following transformation relationships: in is the value of the direction vector v in the x, y, and z directions; is the polar coordinate unit vector in the camera coordinate system.

5. The construction site work center estimation method according to claim 1, 2 or 3, characterized in that: The specific process of step S05 is: S51, assigning different clustering weights to different targets based on prior information; S52. Cluster the weighted polar coordinate unit vectors using a mean shift method to obtain cluster centers; wherein the kernel function used in the mean shift method is a Gaussian function.

6. A construction site work center estimation system, configured to execute the steps of the construction site work center estimation method according to any one of claims 1 to 5, characterized in that: include: A first program module is used to acquire an image of the construction site through an image acquisition unit, and calibrate the image acquisition unit to obtain calibration parameters; A second program module is used to detect key targets in the image of the construction site, wherein the key targets include a person's head, a hard hat, a utility pole, and a safety fence; The third program module is used to calculate the image center of gravity of each key target in the current field of view to obtain an image center of gravity set; The fourth program module is used to traverse the image centroid set, dedistort the image centroid of each key target according to the calibration parameters to obtain normalized coordinates, calculate the direction vector of the normalized coordinates in the image acquisition unit coordinate system, and then convert the direction vector into a unit polar coordinate vector; The fifth program module is used to cluster the unit polar coordinate vectors to obtain the cluster center, which is the construction site work center.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 5.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the computer program is executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 5.

Images