Pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision

Abstract

The invention discloses a pipeline hoisting safety analysis method based on an RBF neural network and binocular vision, and belongs to the field of pipeline hoisting, and the method comprises the following steps: S1, collecting an image of a pipeline hoisting operation site; s2, measuring features in the image according to a binocular vision ranging technology; s3, analyzing the collected image and the measured features, and establishing an RBF neural network model; s4, identifying and judging pipeline hoisting safety behaviors according to the RBF neural network model; and S5, performing monitoring according to the identification and judgment conditions, and performing early warning action, so that the unsafe behavior state of the pipeline hoisting operation site can be monitored in real time, the site risk state can be judged, and an early warning signal can be sent out, thereby replacing an on-site manual patrol monitoring mode, timely finding the safety problem existing in the construction site, reducing the existing potential safety hazard, and improving the safety of the construction site. And the safety of operators is ensured.

Description

Technical Field

[0001] The invention relates to the field of pipeline hoisting, and in particular to a pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision. Background Art

[0002] With the growing global demand for energy, the consumption of traditional fossil fuels such as oil and natural gas has gradually increased, and the number of pipelines for transporting oil and gas has inevitably increased year by year. However, there are many safety hazards at the oil and gas pipeline hoisting site, and the incidence of safety accidents remains high. On the one hand, the on-site workers' active safety protection is not in place, and there are many unsafe behaviors such as not wearing safety helmets and work clothes. On the other hand, during pipeline hoisting operations, the safe distance between the hoisted objects and the workers is often not effectively guaranteed due to communication and operational errors, resulting in serious safety accidents. In the traditional pipeline hoisting site, unsafe behaviors and risk status monitoring and management rely on manual inspections and experience judgments. Inspectors need to observe and monitor in real time, which is not only time-consuming and laborious, but also helpless for the safety risks caused by communication and operational errors in hoisting operations. It cannot effectively guarantee the safety of on-site workers. At the same time, there are problems such as low automation level, large workload and untimely response. Therefore, how to evaluate the safety hazards existing at the pipeline hoisting site has great guiding value for operations. Summary of the invention

[0003] In response to the above problems, the present invention provides a pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision, which can monitor the unsafe behaviors and risk status at the oil and gas pipeline hoisting operation site in advance, reduce the safety hazards at the construction site, and improve the level and efficiency of safety monitoring.

[0004] The technical solution of the present invention is:

[0005] The pipeline hoisting safety analysis method based on RBF neural network and binocular vision includes the following steps:

[0006] S1. Collect images of the pipeline hoisting operation site;

[0007] S2, based on binocular vision ranging technology, the features in the image are measured;

[0008] S3, analyzing the collected images and the measured features, and establishing an RBF neural network model;

[0009] S4, identify and determine pipeline hoisting safety behaviors based on RBF neural network model;

[0010] S5. Monitor and take early warning actions based on the identification and judgment situation.

[0011] In the step S1, visual equipment is used to capture and photograph images of the pipeline hoisting site, and the objects of the images are the on-site workers and the hoisting pipelines.

[0012] In step S1, the collected image is enhanced. The image enhancement method is as follows:

[0013] 1) Image preprocessing: denoising the image;

[0014] 2) Positioning and segmentation: The image after noise reduction is positioned and segmented using an edge-based segmentation method to determine the position, size, direction and other information of the target features in the image;

[0015] 3) Feature extraction: Extract features from the image after positioning and segmentation to obtain feature information such as shape, area, grayscale, texture, and color of the target feature image.

[0016] In step S2, the binocular vision ranging method is as follows:

[0017] 1) Construct a binocular visual ranging system according to the on-site conditions;

[0018] 2) Calculate the target feature information parallax based on the binocular vision ranging system;

[0019] 3) According to the parallax of the target feature information, the target feature information is measured to obtain the real-time shortest distance information.

[0020] In the step 2), the target feature information disparity is calculated using a stereo matching algorithm combined with cost aggregation to obtain an optimal disparity.

[0021] The step S3 is to identify and classify the collected images and the measured features, establish an unsafe behavior model library and a risk distance determination model, and then construct an RBF neural network model.

[0022] The present analysis method also includes an analysis system, which includes a visual information acquisition module, a visual image enhancement module, a binocular vision ranging module, a behavior state recognition module, a monitoring and early warning module and a controller, and the controller is signal-connected to the visual information acquisition module, the visual image enhancement module, the binocular vision ranging module, the behavior state recognition module, and the monitoring and early warning module.

[0023] The beneficial effects of the present invention are:

[0024] This technology can monitor the unsafe behavior status of pipeline hoisting operations in real time, determine the risk status of the site, and issue early warning signals. This replaces the manual inspection and monitoring method on site, timely discovers safety problems at the construction site, reduces existing safety hazards, and ensures the safety of operators. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of a pipeline hoisting safety analysis method based on RBF neural network and binocular vision according to an embodiment of the present invention;

[0026] Figure 2 It is a logic diagram of the pipeline hoisting safety analysis method based on RBF neural network and binocular vision according to an embodiment of the present invention;

[0027] Figure 3 It is a schematic diagram of an RBF neural network of a pipeline hoisting safety analysis method based on an RBF neural network and binocular vision according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0029] Example:

[0030] like Figure 1-Figure 2 As shown in the figure, the pipeline hoisting safety analysis method based on RBF neural network and binocular vision includes the following steps:

[0031] S1. Collect images of the pipeline hoisting operation site;

[0032] Visual equipment is used to capture and shoot images of the pipeline hoisting site. The objects of collection are on-site workers and hoisting pipelines. Generally, a monocular CCD camera is used and installed symmetrically around the pipeline hoisting site.

[0033] Perform image enhancement processing on the collected images. The image enhancement processing method is as follows:

[0034] 1) Image preprocessing: perform noise reduction on the image to avoid the influence of noise on image recognition;

[0035] Noise reduction processing mainly uses Gaussian filtering to reduce noise on images. A filter template is used to scan each pixel in the image. The grayscale value of each pixel is obtained by weighted average of its own pixel value and the pixel value in the neighborhood. The size of the neighborhood and the weight of the pixels in the neighborhood are determined by the filter template. The size of the filter template is 3*3.

[0036] The gradient amplitude and direction of each pixel in the denoised image are used to perform non-maximum suppression on the gradient amplitude graph, suppressing non-edge pixels to 0 and retaining edge pixels with the largest gradient amplitude. Through double threshold processing, two thresholds are set, a low threshold and a high threshold. Points less than the low threshold are set to 0 as false edges, and points greater than the high threshold are set to 1 as strong edges. The purpose is to reduce false edge points and connect strong edges and weak edges into complete edges.

[0037] 2) Positioning and segmentation: The image after noise reduction is positioned and segmented using an edge-based segmentation method to determine the position, size, direction and other information of the target features in the image;

[0038] The pre-processed denoised image has obvious grayscale jumps or discontinuities at the boundaries of different areas. The edge of the target area is located to achieve image segmentation, and the position, size, direction and other information of the target features in the pixel image are determined, which is conducive to subsequent further processing.

[0039] 3) Feature extraction: Extract features from the image after positioning and segmentation to obtain the shape, area, grayscale, texture, color and other feature information of the target feature image;

[0040] The shape, area, grayscale, texture, color and other information that can effectively describe the feature image are extracted from the segmented image, and the same type of targets have a small intra-class distance and different types of targets have a large inter-class distance, so as to determine the target feature information. The purpose of feature extraction is to improve the accuracy and robustness of subsequent target recognition and reduce computational complexity.

[0041] S2, based on binocular vision ranging technology, the features in the image are measured;

[0042] The specific method is as follows:

[0043] 1) Construct a binocular visual ranging system according to the on-site conditions;

[0044] The monocular CCD camera is calibrated using Zhang Zhengyou calibration method, and the distortion coefficient and internal and external parameters of the monocular CCD camera are analyzed. The analyzed monocular CCD camera parameters are used to perform stereo correction on every two adjacent monocular CCD cameras installed on site to construct 2N binocular vision ranging systems.

[0045] Stereo correction uses the Bouguet algorithm to give the rotation transformation matrix and translation transformation matrix between adjacent CCD cameras. The two-dimensional points on the image are projected into a three-dimensional coordinate system using the calibrated distortion coefficient and internal and external parameter matrix of each monocular CCD camera. The reprojection matrix is ​​as follows:

[0046]

[0047] Where Q represents the reprojection matrix in the three-dimensional coordinate system, T x represents the translation transformation matrix in the left figure, c x represents the x-coordinate in the left figure, c′ x represents the x coordinate in the right figure, c y represents the y coordinate in the left figure, and f represents the focal length of the camera.

[0048] Projecting a known two-dimensional homogeneous point into three dimensions, the reprojection matrix is:

[0049]

[0050] Where d represents the parallax of the two associated cameras, X represents the corrected x-coordinate, Y represents the corrected y-coordinate, Z represents the corrected z-coordinate, and W represents the corrected w-coordinate.

[0051] The corrected coordinates are (X / W, Y / W, Z / W). The correction lookup mapping table of the left and right views and the mapping function of any image to image are calculated through the correction coordinate function. Then, corresponding to the pixel position on the target feature image of each two adjacent monocular CCD cameras, the floating point position on the corresponding source image is found, and then the new mapping relationship function is interpolated using the integer values ​​of the surrounding source pixels, thereby constructing 2N binocular vision ranging systems for the pipeline lifting operation site.

[0052] The purpose of stereo correction is to minimize the number of reprojections of each image and maximize the observation area, while ensuring that the optical axes of the left and right cameras in each binocular ranging system are parallel and the image rows are aligned.

[0053] 2) Calculate the target feature information parallax based on the binocular vision ranging system;

[0054] Perform stereo matching on the enhanced images, obtain the three-dimensional coordinate relationship between the left and right images of the binocular vision ranging device, and calculate the parallax of the target feature information;

[0055] The stereo matching algorithm is as follows:

[0056] First, when calculating the initial matching cost through the AD algorithm, an upper limit is set to reduce the impact of mismatching in the superposition process:

[0057]

[0058] Among them, C represents the matching cost, p and q represent the two points in the left and right figures, T represents the set upper limit, and I L Represents the grayscale value in the left image, I R Represents the grayscale value in the right image.

[0059] Then, through cost aggregation calculation and adding visual smoothness constraints, an approximate global energy function is established:

[0060]

[0061] Among them, E represents the global energy function, the first term is the data term, which represents the sum of the matching costs of all pixels based on the disparity d, and the second and third terms are smoothing terms, where P1 and P2 are constants, representing the matching point p in N respectively.P Penalty factor for parallax variation within the domain.

[0062] Solving the global minimum of the energy function on a two-dimensional image is an NP-complete problem. The smooth constraints of the two-dimensional case are approximated by merging one-dimensional smooth constraints in multiple directions:

[0063]

[0064] Among them, S represents the sum of the matching costs on the scanning path, L r Represents the cost energy function along the scan path r.

[0065] Finally, the WTA optimization method is used to minimize the above formula to calculate the final disparity d, that is, the disparity value corresponding to the minimum cost path of the pixel point:

[0066] d=min d S(p, d).

[0067] 3) According to the parallax of the target feature information, the target feature information is measured to obtain the real-time shortest distance information;

[0068] The specific method is as follows:

[0069] Based on the disparity value, the distance information from each target feature to the imaging plane of the binocular ranging system is solved:

[0070]

[0071] Among them, Z represents the imaging distance, b represents the distance between binocular ranging cameras, f represents the focal length of the camera, and d represents the parallax;

[0072] Angle compensation algorithm is used to solve the problem of angle deviation in distance information measurement of target features at the pipeline hoisting operation site due to the limited number of on-site binocular vision systems. The pixel positions of the hoisting pipeline and each on-site operator are proposed, and the offset angle in each binocular ranging system is calculated. Finally, the Dijkstra algorithm is used to match and calculate the shortest distance between the contour information of the personnel at the pipeline hoisting operation site and the contour information of the hoisting pipeline after angle compensation correction, so as to obtain the real-time shortest distance information, including the shortest lateral distance X D and the shortest longitudinal distance Y D .

[0073] S3, analyzing the collected images and the measured features, and establishing an RBF neural network model;

[0074] The collected images and measured features are identified and classified, and an unsafe behavior model library and a risk distance determination model are established, and then an RBF neural network model is constructed.

[0075] like Figure 3 As shown in Figure 1, the RBF neural network consists of an input layer, a hidden layer, and an output layer.

[0076] The steps to establish the RBF neural network model are as follows:

[0077] First, the input layer sends the input signal to the hidden layer. The input node of the input layer is x = [x1 x2...x n ] T , the input nodes convert the input data of the neural network to the hidden layer, and the number of neurons in the input layer nodes is equal to the dimension of the input data vector of the neural network.

[0078] Secondly, the low-dimensional input is mapped to a high-dimensional space through the radial basis function in the hidden layer and curve fitting is performed. The radial basis function used in this embodiment is a Gaussian function, as follows:

[0079]

[0080] Among them, h represents the output value of the hidden layer, j represents the jth neuron in the hidden layer, and the center vector value of the jth neuron is calculated by c j Indicates that c j =[c j1 c j1 …c j1 ], i = 1, 2, ..., n; b j represents the central vector width of the Gaussian basis function, b j =[b1 b2...b j ] T , b j >0,j=1,2,...,m.

[0081] The weights between the hidden layer and the output layer can be expressed as:

[0082] w=[w1 w2...w m ] T

[0083] Among them, w represents the weight coefficient between the hidden layer and the output layer.

[0084] Finally, the output of the output layer can be expressed as:

[0085] y m =w1h1+w2h2+…+w m h m

[0086] Where y represents the predicted output of the RBF neural network.

[0087] S4, identify and determine pipeline hoisting safety behaviors based on RBF neural network model;

[0088] Based on the target feature image after image enhancement processing, the unsafe behaviors at the pipeline lifting operation site are identified and classified into the subordinate sub-model library; through the risk distance judgment model, the real-time shortest distance information of the target feature is judged.

[0089] S5. Monitor and take early warning actions based on the identification and judgment situation.

[0090] The sub-model library of the unsafe behavior model library includes a primary risk model library and a secondary risk model library. When the RBF neural network model recognizes that the target feature image is in the primary risk model library, it is determined that the current state is a primary risk. When the target feature image is recognized to be in the secondary risk model library, it is determined that the current state is a secondary risk. The specific interpretation method is as follows:

[0091] The real-time distance information is compared with the preset risk distance judgment threshold horizontal distance X max , longitudinal distance Y max For comparison, the real-time lateral distance X D >X max When X D ≤X max When the real-time longitudinal distance Y D ≥Y max When Y D <Y max , it is determined that the current status is a level 2 risk.

[0092] When the RBF neural network model detects that the on-site risk status is a level one risk, an early warning signal is issued on site. The early warning signal includes an audible and visual signal emitted by an alarm installed at the work site. If the risk hazards are eliminated within the preset early warning time, the risk status is reduced to a safe status and the early warning is stopped. Otherwise, the risk status rises to a level two risk. When the risk status is a level two risk, the system issues a shutdown command and sends a level two risk signal to the on-site management personnel at the same time. The machine can be restarted only after the on-site management personnel eliminate the risk hazards. The analysis method also includes an analysis system, which includes a visual information acquisition module, a visual image enhancement module, a binocular vision ranging module, a behavior state recognition module, a monitoring and early warning module, and a controller. The controller is connected to the visual information acquisition module, the visual image enhancement module, the binocular vision ranging module, the behavior state recognition module, and the monitoring and early warning module signals.

[0093] Visual information acquisition module, used to capture and shoot images of pipeline hoisting operation sites;

[0094] A visual image enhancement module is used to process the collected visual image information to obtain an enhanced image;

[0095] Binocular vision ranging module, which uses binocular vision ranging technology to measure the distance of target features in the captured image in real time;

[0096] The behavior state recognition module recognizes and determines the target feature image and distance information based on the RBF neural network model;

[0097] The monitoring and early warning module is used to automatically monitor and warn of detected unsafe behavior states.

[0098] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A pipeline hoisting safety analysis method based on RBF neural network and binocular vision is characterized by: The steps include: S1. Collect images of the pipeline hoisting operation site; S2, based on binocular vision ranging technology, the features in the image are measured; S3, analyzing the collected images and the measured features, and establishing an RBF neural network model; S4, identify and determine pipeline hoisting safety behaviors based on RBF neural network model; S5. Monitor and take early warning actions based on the identification and judgment situation.

2. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to claim 1 is characterized in that: In the step S1, visual equipment is used to capture and photograph images of the pipeline hoisting site, and the objects of the images are the on-site workers and the hoisting pipelines.

3. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to claim 1 is characterized in that: In step S1, the collected image is enhanced. The image enhancement method is as follows: 1) Image preprocessing: denoising the image; 2) Positioning and segmentation: The image after noise reduction is positioned and segmented using an edge-based segmentation method to determine the position, size, direction and other information of the target features in the image; 3) Feature extraction: Extract features from the image after positioning and segmentation to obtain feature information such as shape, area, grayscale, texture, and color of the target feature image.

4. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to claim 1 is characterized in that: In step S2, the binocular vision ranging method is as follows: 1) Construct a binocular visual ranging system according to the on-site conditions; 2) Calculate the target feature information parallax based on the binocular vision ranging system; 3) According to the parallax of the target feature information, the target feature information is measured to obtain the real-time shortest distance information.

5. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to claim 4 is characterized in that: In the step 2), the target feature information disparity is calculated using a stereo matching algorithm combined with cost aggregation to obtain an optimal disparity.

6. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to claim 1 is characterized in that: The step S3 is to identify and classify the collected images and the measured features, establish an unsafe behavior model library and a risk distance determination model, and then construct an RBF neural network model.

7. The pipeline hoisting safety behavior analysis method based on RBF neural network and binocular vision according to any one of claims 1 to 6, characterized in that: The present analysis method also includes an analysis system, which includes a visual information acquisition module, a visual image enhancement module, a binocular vision ranging module, a behavior state recognition module, a monitoring and early warning module and a controller, and the controller is signal-connected to the visual information acquisition module, the visual image enhancement module, the binocular vision ranging module, the behavior state recognition module, and the monitoring and early warning module.

Images