Steel structure hoisting posture anomaly detection device and method based on improved Unet semantic segmentation

Abstract

The present invention provides a steel structure hoisting posture abnormality detection device and method based on improved Unet semantic segmentation, the hoisting equipment includes a chassis and a boom; the camera of the image acquisition unit is set to obtain a real-time image of the chassis posture and a real-time image of the boom posture; the image processing unit includes a memory and a comparator, the memory is set to store a set range image of the chassis posture and a set range image of the boom posture, the comparator is set to receive the real-time image of the chassis posture and compare it with the set range image to determine whether the chassis posture exceeds the set range, receive the real-time image of the boom posture and compare it with the set range image to determine whether the boom posture exceeds the set range; the alarm unit is set to send an alarm signal when the chassis posture or the boom posture exceeds the set range. The detection method is deployed on the image processing unit and is divided into three steps. First, the semantic segmentation method based on deep learning is used to segment the boom skeleton and the key points of the skeleton connection, and then the skeleton information of the boom is extracted using a traditional visual algorithm, the key points at both ends of the skeleton are used as vertices, and the angles between different skeletons are calculated, so as to avoid the problem of overturning caused by the posture of the hoisting equipment exceeding the reasonable range, and improve the hoisting construction level and safety level.

Description

Technical Field

[0001] The present invention relates to the field of building construction technology, and in particular to a steel structure hoisting posture abnormality detection device and method based on image semantic segmentation. Background Art

[0002] The steel structure of large buildings is divided into frame structure and large-span truss structure. The main components include steel columns, steel beams, and roof trusses. The span of large buildings is often more than 100 meters. After the overall lifting of the frame structure is completed, truss sliding construction is required.

[0003] The current hoisting and sliding construction mostly uses crawler hoists. Crawler hoists use chassis walking, and the boom swings and retracts independently. The boom does not have a separate power source, and obtains power from the main pump of the main engine through a hydraulic oil pipe. The chassis and boom drive can be linked or single-acting, and can be freely switched in different operating modes. During the construction process, the crawler hoist can be installed while walking, which is highly efficient. However, in the existing steel structure installation process, the driver of the hoisting equipment judges the status of the hoisting equipment based on experience. If there is a slight mistake, once the posture of the boom and hoisting equipment exceeds the reasonable range, safety accidents are likely to occur. In the face of the development of automation and intelligence of devices, designing a method for hoisting posture recognition for large buildings can effectively increase the level of automation and intelligence of hoisting, the convenience of construction organization and management, and safety. Summary of the invention

[0004] In view of the problem that during the existing steel structure installation process, the driver judges the status of the lifting equipment based on experience, and the boom and the lifting equipment are prone to exceed the reasonable range and overturning accidents are prone to occur, the present invention provides a steel structure lifting equipment and method based on posture image recognition monitoring, so as to achieve real-time monitoring of the posture of the lifting equipment, so as to avoid the posture of the lifting equipment exceeding the reasonable range and not being discovered in time, causing collision and overturning problems.

[0005] In order to achieve the above-mentioned purpose, in the first aspect, the present invention provides a steel structure hoisting equipment based on posture image recognition monitoring. The hoisting equipment includes a chassis and a boom, and also includes an image acquisition unit, an image processing unit and an alarm unit, wherein the image acquisition unit is connected to the input end of the image processing unit, and the alarm unit is connected to the output end of the image processing unit; the image acquisition unit includes a camera, and the camera is configured to acquire a real-time image of the chassis posture and a real-time image of the boom posture; the image processing unit includes a semantic segmentation unit and a visual algorithm unit, and the alarm unit includes a comparison alarm unit, and the alarm unit is configured to send an alarm signal when the chassis posture or the boom posture exceeds a set range.

[0006] The specific implementation steps are as follows:

[0007] Step 1: The image acquisition unit collects the complete image of the lifting equipment and performs detailed semantic segmentation and annotation of the boom and each jib connection point;

[0008] Step 1.1: Use the camera to capture images of the hoisting equipment during lifting.

[0009] Step 1.2: Annotate the acquired images, marking the edge contours of the boom and the connection between the booms and key point information.

[0010] Step 2: The image processing unit uses the labeled data set to train the semantic segmentation unet model, so that the unet model can accurately obtain the edge information of the boom and the boom connection and the key point position information of the boom connection;

[0011] Step 2.1: Put the picture of the lifting equipment captured by the camera in step 1.1 and the label file annotated in step 1.2 into the network of the semantic segmentation unet model for training.

[0012] Step 2.2: Use the cross entropy loss function to optimize the semantic segmentation task, use L1loss to optimize the key point regression task, and save the trained model after the model converges.

[0013] Step 3: Use the annotated video and model configuration to train the unet semantic segmentation model, which can accurately detect the key information of the hoisting;

[0014] Step 3.1: Obtain the edge information of the boom and the boom connection points of the lifting equipment through the semantic segmentation head.

[0015] Step 3.2: Obtain the key point information of the boom through the key point regression task head.

[0016] Step 4: Use OpenCV visual algorithm to extract detailed skeleton information of the crane image and the location information of each key point.

[0017] Step 4.1: Through binary image refinement, the skeleton information of the crane arm is extracted and the straight line information of the skeleton is fitted.

[0018] Step 4.2: Extract the key point coordinates predicted in the regression task and only keep the points in the boom connection point area, where the boom connection point is obtained in step 3.1;

[0019] Step 5: Use the key points at both ends of the skeleton extracted in 4.2 as a pair of points to determine the straight line, and repeat this process to calculate the angle between the straight lines.

[0020] Step 6: If any suspicious operation suspected of violation occurs, the alarm unit will alarm the illegal action and save a screenshot. Otherwise, repeat steps 3, 4 and 5.

[0021] Optionally, the image processing unit further comprises a semantic segmentation unit, and the image processor is configured to perform contour processing, signal amplification, dot matrixing and obtain dot matrixed contours of the real-time image of the chassis posture and the real-time image of the boom posture.

[0022] Optionally, the comparator is configured to compare and identify the real-time image of the chassis posture and the real-time image of the boom posture with the set range image respectively.

[0023] Optionally, the memory stores a dotted outline of a set range of chassis posture and a dotted outline of a set range of boom posture, and the comparator is configured to compare and identify the dotted outline of the chassis posture and the dotted outline of the boom posture with the dotted outline of the set range respectively.

[0024] Optionally, the steel structure lifting equipment based on posture image recognition monitoring also includes a posture control unit, which is connected to the output end of the image processing unit. The posture control unit is used to control the chassis posture and the boom posture. The chassis posture control includes chassis position control, and the boom posture control includes boom rotation angle control and height lifting control.

[0025] In order to achieve the above object, in a second aspect, the present invention provides a steel structure hoisting method based on posture image recognition monitoring, comprising:

[0026] S1. Image acquisition: The image acquisition unit acquires a real-time image of the chassis posture of the crawler crane and a real-time image of the boom posture;

[0027] S2. Image contour acquisition: The image processor of the image processing unit processes the real-time image of the chassis posture and the real-time image of the boom posture input by the image acquisition unit to obtain the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture;

[0028] S3. Image key point regression acquisition: The key points in the image are acquired through the key point regression head in the semantic segmentation unit;

[0029] S4. Image contour comparison processing: The traditional vision unit processes the contour of the real-time image obtained by the bottom semantic segmentation unit to obtain more accurate skeleton information by calculating the angle between the booms and comparing the contour with the set range;

[0030] S5. Real-time alarm processing: If the contour of the real-time image of the chassis posture exceeds the contour of the set range or the contour of the real-time image of the boom posture exceeds the contour of the set range, the alarm is triggered; if the contour of the real-time image of the chassis posture exceeds the contour of the set range and the contour of the real-time image of the boom posture does not exceed the set range, return to step S1 and continue to acquire the real-time image.

[0031] Optionally, the steel structure hoisting method based on posture image recognition monitoring also includes:

[0032] Image contour dot matrix processing: The image processor of the image processing unit processes the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture into dot matrix;

[0033] Dot matrix image comparison processing: The comparator compares the outline of the dot matrix chassis posture real-time image and the outline of the boom posture real-time image with the set range.

[0034] The present invention provides a steel structure hoisting method based on posture image recognition monitoring. During the movement of the chassis and the boom, the camera of the image acquisition unit acquires the real-time image of the chassis posture and the real-time image of the boom posture. The image processing unit receives the real-time image of the chassis posture and compares it with the set range image to determine whether the chassis posture exceeds the set range, receives the real-time image of the boom posture and compares it with the set range image to determine whether the boom posture exceeds the set range; when the chassis posture or the boom posture exceeds the set range, the alarm of the alarm unit sends an alarm signal. Thus, the posture of the hoisting equipment is monitored in real time and controlled within a reasonable range, avoiding the problem of overturning due to the posture of the hoisting equipment exceeding the reasonable range, and improving the hoisting construction level and safety level. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. The accompanying drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0036] Figure 1 It is a system configuration diagram of the first preferred embodiment of the steel structure hoisting equipment based on posture image recognition and monitoring provided by the present invention.

[0037] Figure 2 It is a structural schematic diagram of the image acquisition unit.

[0038] Figure 3 It is a schematic diagram of the improved Unet semantic segmentation structure.

[0039] Figure 4It is a system configuration diagram of the second preferred embodiment of the steel structure hoisting equipment based on posture image recognition and monitoring provided by the present invention.

[0040] Figure 5 It is a flowchart of a preferred embodiment of the steel structure hoisting method based on posture image recognition and monitoring provided by the present invention.

[0041] In the figure, 1 is an image acquisition unit, 11 is a first camera, 12 is a second camera, 13 is a third camera; 2 is an image processing unit, 21 is a semantic segmentation processing unit, 22 is a traditional visual algorithm processing unit; 3 is an alarm unit, 31 is a comparison and reporting unit; 4 is a posture control unit. DETAILED DESCRIPTION

[0042] The steel structure of large buildings is divided into frame structure and large-span truss structure. The main components include steel columns, steel beams, and roof trusses. The span of large buildings is often more than 100 meters. After the overall hoisting of the frame structure is completed, the truss sliding construction needs to be carried out. The current hoisting and sliding construction mostly use crawler hoists. During the construction process, crawler hoists can be installed while walking, which is highly efficient. However, in the existing steel structure installation process, the driver of the hoisting equipment judges the status of the hoisting equipment based on experience. If you are not careful, once the posture of the boom and hoisting equipment exceeds the reasonable range, it is easy to overturn.

[0043] To this end, the present invention provides a steel structure hoisting device and method based on posture image recognition monitoring. The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present invention and its application or use.

[0044] Figure 1 1 is a system configuration diagram of the first preferred embodiment of the steel structure hoisting equipment based on posture image recognition monitoring provided by the present invention. Figure 1As shown, in some preferred embodiments, the present invention provides a steel structure hoisting equipment based on posture image recognition monitoring, the hoisting equipment includes a chassis and a boom, an image acquisition unit 1, an image processing unit 2 and a comparison alarm unit 3, the image acquisition unit 1 is connected to the input end of the image processing unit 2, and the alarm unit 3 is connected to the output end of the image processing unit 2; the image acquisition unit 1 includes a camera, and the camera is configured to acquire a real-time image of the chassis posture and a real-time image of the boom posture; the image processing unit 2 includes a semantic segmentation unit 21 and a traditional visual algorithm unit 22, the semantic segmentation unit 21 is configured to segment and acquire the dot edge information of the boom and key points in real time, and the traditional visual algorithm unit 22 is configured to receive a real-time image of the chassis posture and compare it with a set range image to determine whether the chassis posture exceeds the set range, receive a real-time image of the boom posture and compare it with a set range image to determine whether the boom posture exceeds the set range;

[0045] The comparison alarm unit 3 is configured such that when the chassis posture or the boom posture exceeds a set range, the alarm 31 sends out an alarm signal.

[0046] Compared with the prior art, the steel structure hoisting equipment based on posture image recognition monitoring provided by the present invention comprises a chassis and a boom; the camera of the image acquisition unit is configured to acquire a real-time image of the chassis posture and a real-time image of the boom posture; the image processing unit comprises a memory and a comparator, the memory is configured to store a set range image of the chassis posture and a set range image of the boom posture, the comparator is configured to receive a real-time image of the chassis posture and compare it with a set range image to determine whether the chassis posture exceeds the set range, receive a real-time image of the boom posture and compare it with a set range image to determine whether the boom posture exceeds the set range; the alarm unit is configured to send an alarm signal when the chassis posture or the boom posture exceeds the set range. The posture of the hoisting equipment can be monitored to avoid the problem of overturning caused by the posture of the hoisting equipment exceeding a reasonable range, and improve the level of hoisting construction and safety.

[0047] Figure 2 It is a schematic diagram of the structure of the image acquisition unit. Figure 2 As shown, the cameras are placed at the positions of the first camera 11, the second camera 12, and the third camera 13, and the images detected by the three cameras are calibrated and spliced ​​by homography matrix transformation, so as to obtain a complete image of the hoisting equipment. Preferably, the camera can be a video camera, a visual detector, or an ultrasonic detector.

[0048] Figure 3It is an improved version of the semantic segmentation network provided by the present invention. In addition to the traditional Unet structure, it has an additional key point prediction head, which makes it a multi-task network. It can predict the image coordinates of the key points between the boom structures while detecting the boom contour, so as to achieve direct and accurate pairing of the key points. First, given its input image, it performs four convolution operations on it, continuously extracts features, and then performs bilinear interpolation upsampling;

[0049] (1)

[0050] (X0, Y0) and (X1, Y1) are the coordinates of known image points.

[0051] f(X0,Y0),f(X0,x_0,Y0),f(X1,Y0),f(X0,Y1),f(X1,Y1) are the function values ​​corresponding to these image points.

[0052] (x,y) are the coordinates of the image point that needs to be interpolated.

[0053] After upsampling, the sampled feature map is spliced ​​with the previous layer to achieve the purpose of feature fusion. After several operations, the mask of the boom is obtained.

[0054] like Figure 3 As shown in Figure 1, in order to predict the key points, the sampled feature map is reduced in dimension after the first downsampling, and a fully connected network is used to form a regression task to predict the image coordinates of the key points at the boom connection.

[0055] like Figure 3 As shown in Figure 2, in order to train a multi-task network, different loss functions need to be used to make the network converge. The regression task of the key point coordinates is shown in Formula 2:

[0056] (2)

[0057] In formula 2, x, y are predicted values, yi, xi are true values, and the loss function is calculated by directly interpolating the two. For the semantic segmentation network, the loss function is shown in formula 3

[0058] (3)

[0059] L represents the loss value. is the true label (0 or 1). is the probability predicted by the model. n is the sample size. The final loss function of the entire network is shown in formula (4):

[0060] (4)

[0061] Figure 4 1 is a system configuration diagram of a second preferred embodiment of a steel structure hoisting device based on posture image recognition monitoring provided by the present invention. Figure 4 As shown, in some preferred embodiments, the image processing unit 2 also includes an image processor 23, and the image processor 23 is configured to process the contours of the real-time image of the chassis posture and the real-time image of the boom posture, amplify the signal, dot-matrix, and obtain the dot-matrix contours. The comparator 21 is configured to compare and identify the real-time image of the chassis posture and the real-time image of the boom posture with the set range image respectively. The memory 21 stores the dot-matrix contours of the set range of the chassis posture and the dot-matrix contours of the set range of the boom posture, and the comparator 21 is configured to compare and identify the dot-matrix contours of the chassis posture and the dot-matrix contours of the boom posture with the dot-matrix contours of the set range respectively. In this embodiment, the image processor 23 performs contour processing, signal amplification, rasterization, and acquisition of the rasterized contour of the real-time image of the chassis posture and the real-time image of the boom posture, so that the information of the chassis posture and the boom posture can be obtained more accurately, and invalid information in image recognition can be eliminated, thereby reducing the amount of calculation in the image recognition process, thereby reducing the amount of calculation of the image processing unit, improving the processing speed, and being able to more quickly discover problems where the chassis posture and the boom posture exceed the set range, thereby more promptly avoiding the occurrence of overturning problems and improving the safety of equipment operation.

[0062] like Figure 4 As shown, in some preferred embodiments, the steel structure hoisting equipment based on semantic segmentation monitoring also includes a posture control unit 4, which is connected to the output end of the image processing unit 2. The posture control unit 4 is used to control the chassis posture and the boom posture. The chassis posture control includes the chassis position control, and the boom posture control includes the boom rotation angle control and the height lifting control. In this embodiment, the device is connected to the posture control unit 4 by the image processing unit 2, and whether the chassis posture and the boom posture exceed the set range is obtained in time. When the chassis posture and the boom posture exceed the set range, the posture control unit 4 is used to adjust the chassis posture and the boom posture to within the set range in time, so as to actively avoid the occurrence of overturning problems, without the need for driver operation, so that the operation level and safety of the equipment are effectively improved.

[0063] Figure 5 1 is a flowchart of a preferred embodiment of the steel structure hoisting method based on posture image recognition and monitoring provided by the present invention. Figure 5 As shown, in some preferred embodiments, the present invention provides a steel structure hoisting method based on posture image recognition monitoring, comprising:

[0064] S1. Image acquisition: The image acquisition unit 1 acquires a real-time image of the chassis posture of the lifting equipment and a real-time image of the boom posture;

[0065] S2. Image contour processing: The semantic segmentation unit 21 of the image processing unit 2 processes the real-time image of the chassis posture and the real-time image of the boom posture input by the image acquisition unit 1, and obtains the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture and the key point information on both sides of the boom;

[0066] S3. Traditional visual algorithm processing: The traditional visual algorithm unit 22 extracts features from the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture to obtain more detailed and accurate boom skeleton information;

[0067] S4. Comparison alarm unit: using the key points on both sides of the boom as vertices, calculating the angle between the booms to determine the posture of the lifting equipment, if the contour of the real-time image of the chassis posture exceeds the contour of the set range or the contour of the real-time image of the boom posture exceeds the contour of the set range, the alarm 31 is triggered;

[0068] S5: If the boom posture exceeds the set range, an alarm is triggered; otherwise, the process returns to step S1 to continue acquiring real-time images.

[0069] Compared with the prior art, the steel structure hoisting method based on posture image recognition monitoring provided by the present invention, during the movement of the chassis and the boom, the camera of the image acquisition unit acquires the real-time image of the chassis posture and the real-time image of the boom posture, the image processing unit receives the real-time image of the chassis posture and compares it with the set range image to determine whether the chassis posture exceeds the set range, receives the real-time image of the boom posture and compares it with the set range image to determine whether the boom posture exceeds the set range; when the chassis posture or the boom posture exceeds the set range, the alarm of the alarm unit sends an alarm signal. Thus, the posture of the hoisting equipment is monitored in real time and controlled within a reasonable range, avoiding the problem of overturning due to the posture of the hoisting equipment exceeding the reasonable range, and improving the hoisting construction level and safety level.

[0070] Compared with the prior art, the steel structure hoisting posture abnormality detection device and method based on image semantic segmentation provided by the present invention, during the movement of the chassis and the boom, the camera of the image acquisition unit acquires the real-time image of the chassis posture and the real-time image of the boom posture, the image processing unit receives the real-time image of the chassis posture and compares it with the set range image to determine whether the chassis posture exceeds the set range, receives the real-time image of the boom posture and compares it with the set range image to determine whether the boom posture exceeds the set range; when the chassis posture or the boom posture exceeds the set range, the alarm of the alarm unit sends an alarm signal. Thus, the posture of the hoisting equipment is monitored in real time and controlled within a reasonable range, avoiding the problem of overturning due to the posture of the hoisting equipment exceeding the reasonable range, and improving the level of hoisting construction and safety.

Claims

1. A steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation, the hoisting equipment includes a chassis and a boom, and is characterized by: It comprises an image acquisition unit (1), an image processing unit (2) and an alarm unit (3), wherein the image acquisition unit (1) is connected to an input end of the image processing unit (2), and the alarm unit (3) is connected to an output end of the image processing unit (2); The image acquisition unit (1) comprises a camera, the camera being configured to acquire a real-time image of the chassis posture and a real-time image of the boom posture; the image processing unit (2) comprises a semantic segmentation unit (21) and a traditional visual algorithm unit (22), the semantic segmentation unit (21) being configured to segment and acquire dot matrix information of the boom and key points in real time, the traditional visual algorithm unit (22) being configured to receive a real-time image of the chassis posture and compare it with a set range image to determine whether the chassis posture exceeds the set range, and to receive a real-time image of the boom posture and compare it with a set range image to determine whether the boom posture exceeds the set range; The alarm unit (3) comprises a comparison alarm unit (31), and the comparison alarm unit (3) is configured such that when the chassis posture or the boom posture exceeds a set range, the alarm (31) sends out an alarm signal.

2. According to claim 1, the steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation is characterized by: The specific implementation steps of the device are as follows: Step 1: The image acquisition unit collects the complete image of the lifting equipment and performs detailed semantic segmentation and annotation of the boom and each jib connection point; Step 1.1: Capture the image of the hoisting equipment during lifting through the camera; Step 1.2: Annotate the acquired images, marking the edge contours of the boom and the connection between the booms and key point information; Step 2: The image processing unit uses the labeled data set to train the semantic segmentation unet model, so that the unet model can accurately obtain the edge information of the boom and the boom connection and the key point position information of the boom connection; Step 2.1: Put the pictures of the lifting equipment captured by the camera in step 1.1 and the label files annotated in step 1.2 into the network of the semantic segmentation unet model for training; Step 2.2: Use the cross entropy loss function to optimize the semantic segmentation task, use L1loss to optimize the key point regression task, and save the trained model after the semantic segmentation unet model converges; Step 3: Use the annotated video and model configuration to train the unet semantic segmentation model, which can accurately detect the key information of the hoisting; Step 3.1: Obtain edge information of the boom and boom connection points of the lifting equipment through the semantic segmentation head; Step 3.2: Obtain the key point information of the boom through the key point regression task head; Step 4: Use OpenCV visual algorithm to extract detailed skeleton information of the crane image and the location information of each key point; Step 4.1: Extract the skeleton information of the crane arm through binary image thinning, and fit the straight line information of the skeleton; Step 4.2: Extract the key point coordinates predicted in the regression task and only keep the points in the boom connection point area, where the boom connection point is obtained in step 3.1; Step 5: Use the key points at both ends of the skeleton extracted in 4.2 as a pair of points to determine the straight line, and repeat this process to calculate the angle between the straight lines; Step 6: If any suspicious operation suspected of violation occurs, the alarm unit will alarm the illegal action and save a screenshot. Otherwise, repeat steps 3, 4 and 5.

3. According to claim 1, the steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation is characterized by: The image processing unit further comprises a semantic segmentation unit, and the image processor is configured to perform contour processing, signal amplification, dot matrixing and obtain dot matrixed contours of the real-time image of the chassis posture and the real-time image of the boom posture.

4. According to claim 1, the steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation is characterized by: The comparator is configured to compare and identify the real-time image of the chassis posture and the real-time image of the boom posture with the set range image respectively.

5. The steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation according to claim 1 is characterized by: The memory stores a dotted outline of a set range of chassis posture and a dotted outline of a set range of boom posture, and the comparator is configured to compare and identify the dotted outline of the chassis posture and the dotted outline of the boom posture with the dotted outline of the set range respectively.

6. The steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation according to claim 1 is characterized by: The steel structure lifting equipment based on posture image recognition monitoring also includes a posture control unit, which is connected to the output end of the image processing unit. The posture control unit is used to control the chassis posture and the boom posture. The chassis posture control includes chassis position control, and the boom posture control includes boom rotation angle control and height lifting control.

7. A method for implementing the steel structure hoisting posture anomaly detection device based on improved Unet semantic segmentation as claimed in claim 1, characterized in that: include: S1. Image acquisition: The image acquisition unit acquires a real-time image of the chassis posture of the crawler crane and a real-time image of the boom posture; S2. Image contour acquisition: The image processor of the image processing unit processes the real-time image of the chassis posture and the real-time image of the boom posture input by the image acquisition unit to obtain the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture; S3. Image key point regression acquisition: The key points in the image are acquired through the key point regression head in the semantic segmentation unit; S4. Image contour comparison processing: The traditional vision unit processes the contour of the real-time image obtained by the bottom semantic segmentation unit to obtain more accurate skeleton information by calculating the angle between the booms and comparing the contour with the set range; S5. Real-time alarm processing: If the contour of the real-time image of the chassis posture exceeds the contour of the set range or the contour of the real-time image of the boom posture exceeds the contour of the set range, the alarm is triggered; if the contour of the real-time image of the chassis posture exceeds the contour of the set range and the contour of the real-time image of the boom posture does not exceed the set range, return to step S1 and continue to acquire the real-time image.

8. The method for detecting abnormal posture of steel structure hoisting based on improved Unet semantic segmentation according to claim 7 is characterized in that: Also includes: Image contour dot matrix processing: The image processor of the image processing unit processes the contour of the real-time image of the chassis posture and the contour of the real-time image of the boom posture into dot matrix; Dot matrix image comparison processing: The comparator compares the outline of the dot matrix chassis posture real-time image and the outline of the boom posture real-time image with the set range.

Images