A rail steel plate shield edge detection method and device

Abstract

The present application relates to a kind of edge detection method and device for guide rail steel plate shield, device includes multi-dimensional image acquisition module, edge detection analysis module and control module.Method includes control module control setting in work area monocular camera carries out original image acquisition, edge detection analysis module carries out edge detection by image processing unit, edge image processing unit and edge image analysis unit, and processing result is passed to control module and constitutes feedback loop, control module judges whether to generate warning signal according to detection result information and stops the operation of guide rail steel plate shield, to ensure the safety of numerical control machine tool.The present application establishes edge detection for guide rail steel plate shield in numerical control machine tool, analyzes edge detection result to judge whether shield is in normal operating state by in embedded development board, with the advantages of strong practicality, low cost.

Description

Technical Field

[0001] This invention belongs to the field of industrial automation, specifically a method and device for edge detection of guide rail steel plate protective covers based on monocular vision. It is applied to the automated detection of guide rail steel plate protective covers on industrial conveyor belts, facilitating automated and safe management of industrial workshops. Background Technology

[0002] Currently, in the field of industrial automation, computer vision inspection is an important direction for CNC machine tools. Compared with manual inspection to monitor for abnormal operation of CNC machine tools, computer vision-based detection can more accurately and promptly identify anomalies and can operate continuously 24 hours a day without rest. Because CNC machine tools need to efficiently and quickly process complex and precise parts, guide rail steel plate protective covers play a crucial role. On the one hand, they prevent chips or other sharp objects from entering the machine tool and damaging CNC machine tool parts; on the other hand, they transport parts into the CNC machine tool for subsequent processing. Therefore, visual monitoring of the guide rail steel plate protective covers to determine whether they are in normal operating condition is a very important issue.

[0003] Current methods for monitoring guide rail steel plate protective covers primarily rely on manual inspections. Inspectors periodically check the site, relying on their visual judgment of the protective cover's operational status. This extensive patrol approach is difficult to supervise and evaluate, and feedback is often delayed, severely impacting production efficiency and quality. If problems are not detected promptly, they can lead to guide rail malfunctions and machine tool production halts. Summary of the Invention

[0004] Based on the aforementioned technical problems, an edge detection method and device for a guide rail steel plate protective cover is proposed to solve the problems of low efficiency and difficulty in timely response to faults in the existing manual inspection method. At the same time, it solves the safety management of the workshop and improves the degree of automation.

[0005] The technical means adopted in this invention are as follows: An edge detection method for a guide rail steel plate protective cover, comprising the following steps:

[0006] S1 Dataset Acquisition: Color images of the guide rail steel plate protective cover were acquired using a camera;

[0007] S2 dataset preprocessing: The color images are converted to grayscale, and the grayscale images are labeled to represent the edges of the protective shield; a dataset is constructed based on the labeled grayscale images for training the neural network;

[0008] S3 network model construction: Gradient information is obtained from grayscale images in the dataset using an improved LBP algorithm, and the pixel differences representing the gradient information are input into the ResNet50 network for training to obtain edge images containing shield edge information.

[0009] S4 Edge Detection: Real-time acquisition of protective shield images, and processing of the protective shield images using an improved LBP algorithm. The pixel differences representing gradient information are input into the trained ResNet50 network to obtain an edge image containing protective shield edge information. Then, non-maximum suppression processing is applied to the edge information to obtain a single-pixel edge image.

[0010] The step of obtaining gradient information from grayscale images in the dataset using an improved LBP algorithm includes the following steps:

[0011] For each pixel, it is taken as the center pixel. The pixel difference between the center and the 8 neighboring pixels around it, the pixel difference between any two adjacent pixels in the 8 neighboring pixels around it, and the pixel difference in the radial direction are calculated respectively. At least one of the three differences is used to characterize the gradient information of the image.

[0012] The improved LBP algorithm and the ResNet50 network form the backbone network.

[0013] The ResNet50 network was trained using the Annotator-robust loss function:

[0014]

[0015] Where λ is a hyperparameter used to balance the ratio of positive to negative samples, P(x) is the sigmoid function, and W represents all parameters to be learned; X i This represents the feature vector of the current pixel;

[0016] The mean of the five annotations in the preprocessing of the S2 dataset was calculated, and y was used to calculate the mean. i This means that a new edge probability map is generated, with a range of [0, 1, where 0 indicates that no labeling was performed in any of the 5 instances, and 1 indicates that the pixel was labeled as an edge pixel in all 5 instances.

[0017] Simultaneously, samples with marginal probability values ​​exceeding the threshold η are designated as positive samples Y. + Samples with a marginal probability value less than η are considered negative samples Y. - .

[0018] An edge detection method for a guide rail steel plate protective cover includes an S5 edge judgment: based on the edge coordinate equation of the protective cover obtained from the single pixel edge image, the distance and angle between the edges of each protective plate of the protective cover are determined to determine whether the guide rail steel plate protective cover is in normal operation.

[0019] The S5 edge detection includes the following steps:

[0020] For a single-pixel edge image, edge L is one end of a fixed guide rail steel plate protective cover. A coordinate system is established with the fixed end edge L of the guide rail steel plate protective cover as the y-axis, and the edges L1, L2, L3, ..., L between each piece of the guide rail steel plate protective cover are obtained. n The coordinate equation of the guide rail steel plate protective cover is given by n, where n is the number of protective plates of the guide rail steel plate protective cover.

[0021] A) Obtain edges L1, L2, L3, ..., L n The angle α formed between the guide rail steel plate protective cover and the fixed edge L; if α > δ, then an abnormality is determined and an alarm is triggered; where δ is the allowable error angle; otherwise, normal operation is maintained;

[0022] B) Calculate the edges L, L1, L2, L3, ..., L respectively. n The distances between two adjacent edges are d1, d2, d3, ..., d n And determine whether the following formula is satisfied:

[0023]

[0024] If the condition is met, it is determined that the guide rail steel plate protective cover is abnormal and an alarm is triggered; otherwise, normal operation continues; where ε is the allowable error distance.

[0025] An edge detection device for a guide rail steel plate protective cover includes:

[0026] The multi-dimensional image acquisition module is used to acquire color images of the guide rail steel plate protective cover in the working area through a camera based on the received image acquisition control signal, and output them to the edge detection and analysis module.

[0027] The edge detection and analysis module includes:

[0028] The image processing unit is used to acquire edge information from the color image of the guide rail steel plate protective cover through the backbone network;

[0029] The edge image processing unit performs non-maximum suppression processing on the obtained edge information to obtain a single-pixel edge image;

[0030] The edge image analysis unit is used to determine whether the protective cover is functioning properly based on the edge information in the single-pixel edge image.

[0031] The beneficial effects of this invention are:

[0032] 1. This invention establishes an edge detection method for the protective cover of the guide rail steel plate in CNC machine tools. By analyzing the edge detection results on an embedded development board, it can determine whether the protective cover is in normal operating condition. It has the advantages of strong practicality and low cost.

[0033] 2. The present invention provides an edge detection method and device for a face-guided guide rail steel plate protective cover. Through a multi-dimensional image acquisition module, an edge detection analysis module, and a control module, a computer vision-based edge detection method and device are constructed, enabling real-time monitoring of the operating status of the steel plate protective cover. Simultaneously, when an abnormality is detected in the steel plate protective cover, the control module will issue a warning signal, thereby taking timely protective measures to ensure the safety of the CNC machine tool system. Attached Figure Description

[0034] Figure 1 This is a flowchart illustrating the process of the edge detection device for the guide rail steel plate protective cover of the present invention.

[0035] Figure 2 This is a schematic diagram of an image captured by a monocular camera positioned directly above the device.

[0036] Figure 3 This is a schematic diagram of the edge of the guide rail steel plate protective cover of the present invention;

[0037] Figure 4a This is a schematic diagram of the fault detection results of the guide rail steel plate protective cover of the present invention. Figure 1 ;

[0038] Figure 4b This is a schematic diagram of the fault detection results of the guide rail steel plate protective cover of the present invention. Figure 2 ;

[0039] Figure 5 The diagram shows the principles of the three forms of the LBP algorithm. Detailed Implementation

[0040] The present invention will now be described in further detail with reference to the accompanying drawings.

[0041] An edge detection device for a guide rail steel plate protective cover includes:

[0042] The system includes a multi-dimensional image acquisition module, an edge detection and analysis module, and a control module. The multi-dimensional image acquisition module receives the image acquisition control signal from the control module, acquires the original image of the guide rail steel plate protective cover equipment in the working area, and outputs it to the edge detection and analysis module for image data processing and analysis. The edge detection result is then fed back to the control module for warning control of the guide rail steel plate protective cover equipment.

[0043] The multi-dimensional image acquisition module includes an acquisition bracket set in the working area and a monocular camera mounted thereon, used to acquire the original image information of the guide rail steel plate protective cover.

[0044] The edge detection and analysis module and the control module are implemented on an embedded development board.

[0045] The edge detection and prediction module includes an image processing unit, an edge image processing unit, and an edge image analysis unit;

[0046] The image processing unit acquires edge image information of the original image of the guide rail steel plate protective cover through the backbone network on the embedded development board;

[0047] The edge image processing unit refines the obtained edge image information to obtain a single-pixel edge image of the guide rail steel plate protective cover;

[0048] The edge image analysis unit will determine whether the single-pixel edge information of the guide rail steel plate protective cover is in normal operation and feed the analysis results back to the control module.

[0049] The backbone network includes three improved LBP algorithms connected sequentially and the ResNet50 backbone network.

[0050] The edge image processing unit uses a non-maximum suppression algorithm to refine the edge image and obtain a single-pixel edge image of the guide rail steel plate protective cover.

[0051] The edge image analysis unit calculates the spacing and angle between each edge of the guide rail steel plate protective cover to determine whether the guide rail steel plate protective cover is in normal operating condition.

[0052] The control module includes an acquisition control circuit and a warning control circuit. The acquisition control circuit sends an image acquisition signal to control the monocular camera in the working area to acquire raw image information. The warning control circuit combines the edge detection results to output a warning command to the guide rail steel plate protective cover equipment.

[0053] An edge detection method for a guide rail steel plate protective cover includes the following steps:

[0054] The control module sends acquisition control signals through the acquisition control circuit to control the monocular camera in the working area to acquire the original image of the guide rail steel plate protective cover;

[0055] The edge detection and analysis module acquires the edge information of the steel plate protective cover in the image processing unit based on the received image data, and outputs it to the edge image processing unit; the edge image processing unit obtains a single-pixel edge image through a non-maximum suppression algorithm, and outputs it to the edge image analysis unit; the edge image analysis unit makes a judgment based on the acquired edge information, calculates the spacing and angle between the edges of each protective cover of the guide rail steel plate protective cover, compares it with a standard preset threshold, and outputs the feedback result to the control module;

[0056] Based on the acquired edge detection result signal, the control module outputs a warning signal to the guide rail steel plate protective cover equipment through the warning control circuit to issue a warning when a fault occurs.

[0057] The image processing unit employs three improved LBP differential processing algorithms to capture image gradient information. The processed image pixel differences are then processed using a ResNet50 backbone network to obtain meaningful semantic features, ultimately generating robust and accurate edges.

[0058] The steel plate protective cover tested is based on existing technology, and its physical shape is as follows: Figure 2 As shown, the main component includes the gray area on the left side of the image, which is one side of a steel plate fixed in a designated position, and several long, strip-shaped protective plates. The protective plates are slightly raised in the middle and extend and retract with the guide rail. The main function of the steel plate protective cover is to effectively prevent iron filings and protect the mechanical guide rail.

[0059] An edge detection device for a guide rail steel plate protective cover, such as Figure 1 As shown, it includes: a multi-dimensional image acquisition module, an edge detection and analysis module, and a control module. The multi-dimensional image acquisition module receives image acquisition control signals from the control module and is responsible for acquiring raw images from the monocular camera directly above the guide rail steel plate protective cover equipment. The acquired image data is then output to the edge detection and analysis module for image data processing and analysis. Based on the edge detection results, feedback is given to the control module for warning control of the guide rail steel plate protective cover equipment.

[0060] The installation angle of a monocular camera is as follows Figure 2 As shown, the camera's installation angle should be adjusted so that the captured image completely covers the fully opened guide rail steel plate protective cover. After installation, the camera needs to be calibrated and corrected to calculate its parameters, which will facilitate the calculation of the coordinate equations of each edge during edge detection.

[0061] The edge map effect finally obtained by the edge image processing unit in the edge detection module is shown in the figure below. Figure 3 As shown in the diagram. Edge L is one end of a fixed guide rail steel plate protective cover. A coordinate system is established with edge L as the y-axis, and the edges L1, L2, L3…L… are calculated respectively. n The coordinate equations of .

[0062] The edge image analysis unit in the edge detection module defines two situations as fault conditions, such as... Figures 4a-4b As shown. Among them Figure 4a Calculate the edges L1L2L3…L respectively. nIf the angle α between the guide rail and the fixed edge L is greater than δ, it is determined that the guide rail steel plate protective cover is abnormal, and the detection result signal is transmitted to the control module to issue a warning signal and suspend the operation of the CNC machine tool. Here, δ is a pre-set allowable error angle.

[0063] Figure 4b Calculate the edges LL1L2L3…L respectively. n The distances between d1d2d3…d n If the following formula exists:

[0064]

[0065] If this condition is found, it is determined that the guide rail steel plate protective cover is abnormal, where ε is a pre-set allowable error distance.

[0066] The specific steps for edge inspection of the guide rail steel plate protective cover are as follows:

[0067] S1 Dataset Acquisition: A large number of images of the guide rail steel plate protective cover from different angles were captured using a camera. During shooting, a suitable light source should be selected to ensure the images are as clear and accurate as possible. The resolution of the acquired images is 2560*1920.

[0068] S2 dataset preprocessing: Images were labeled five times using the Geolabel tool. First, the color images were converted to grayscale to reduce the amount of data to be processed. A weighted average method was used to convert the color images to grayscale to obtain a more reasonable grayscale image; the formula is as follows:

[0069] L=R*299 / 1000+G*587 / 1000+B*114 / 1000

[0070] S3 Dataset Partitioning: The dataset is divided into training, testing, and validation sets in a 7:2:1 ratio;

[0071] S4 Network Model Construction: The baseline model uses an improved LBP algorithm and a ResNet50 backbone, followed by non-maximum suppression to obtain single-pixel edge images. Unlike existing edge detection algorithms, image information is not directly fed into the neural network for training. Instead, the improved LBP algorithm is first applied to the image to capture its gradient information. Then, the ResNet network focuses on extracting meaningful semantic features. The pixel differences between the perimeter and the center, the pixel differences in the perimeter loop, and the pixel differences in the radial direction are calculated to obtain the image gradient information. Specifically, the three forms of the improved LBP algorithm are as follows: Figure 5 As shown.

[0072] The processed image data is fed into the ResNet50 network for training. The Annotator-robust loss function is used, and its specific expression is as follows:

[0073]

[0074] Calculate the mean of the 5 labels in S2, and use y i This indicates that a new edge probability map is generated, with a range of [0, 1, where 0 indicates that no annotation was performed in any of the 5 iterations, and 1 indicates that the pixel was labeled as an edge in all 5 iterations. Simultaneously, edge probability values ​​exceeding η are considered as positive samples Y. + Samples with a marginal probability value less than η are considered negative samples Y. - X i This represents the feature vector of the current pixel. In this patent, η is set to 0.5.

[0075] In the above formula, λ is a hyperparameter used to balance the ratio of positive to negative samples. P(x) is the standard sigmoid function, and W represents all the parameters to be learned.

[0076] S5 Network Model Training: Using the network model in the training set S4 obtained from S2, training is performed to obtain an edge image containing edge information. This allows for the training of a model suitable for edge detection of guide rail steel plate protective covers.

[0077] S6 Image Input: Acquire the actual image of the guide rail steel plate protective cover in the actual machine tool from the monocular camera, and the acquisition rules are in accordance with S1;

[0078] S7 Edge Detection: The image acquired in S6 is input into the S4 network model to detect edge information in the image, and then non-maximum suppression processing is applied to obtain a single-pixel edge image. Finally, the image is obtained... Figure 3 As shown, L1L2L3…L n The coordinate equations;

[0079] S8 Edge Detection: Based on the edge coordinate equations generated in S7, calculate the ratio and angle of the edge distance.

[0080] α i >δ

[0081]

[0082] If either of the above two equations is true, a warning signal will be issued, the operation of the CNC machine tool will be suspended, and manual intervention will be required. Here, δ and ε are pre-defined error values.

[0083] This invention discloses an edge detection method and device for a guide rail steel plate protective cover. Through a multi-dimensional image acquisition module, an edge detection analysis module, and a control module, it constructs a computer vision-based edge detection method and device capable of real-time monitoring of the steel plate protective cover's operating status. This invention addresses the problem of low efficiency in manual inspection systems for guide rail steel plate protective covers, and overcomes the safety deficiencies and inability to respond promptly to faults associated with manual inspection. This invention has the advantages of high practicality and low cost.

[0084] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should be considered within the scope of protection of the present invention.

Claims

1. A method for edge detection of a guide rail steel plate protective cover, characterized in that, Includes the following steps: S1 Dataset Acquisition: Color images of the guide rail steel plate protective cover were acquired using a camera; S2 dataset preprocessing: The color images are converted to grayscale, and the grayscale images are labeled to represent the edges of the protective shield; a dataset is constructed based on the labeled grayscale images for training the neural network; S3 network model construction: Gradient information is obtained from grayscale images in the dataset using an improved LBP algorithm, and the pixel differences representing the gradient information are input into the ResNet50 network for training to obtain edge images containing shield edge information. S4 edge detection: Real-time acquisition of protective shield images, and processing of the protective shield images using an improved LBP algorithm. The pixel differences representing gradient information are input into the trained ResNet50 network to obtain an edge image containing the edge information of the protective shield. Then, non-maximum suppression processing is applied to the edge information to obtain a single-pixel edge image. The step of obtaining gradient information from grayscale images in the dataset using an improved LBP algorithm includes the following steps: For each pixel, take it as the center pixel, calculate the pixel difference between the 8 neighboring pixels around it and the center pixel, the pixel difference between every two adjacent pixels in the 8 neighboring pixels around it, and the pixel difference in the radial direction. Use at least one of the three differences to characterize the gradient information of the image. It also includes S5 edge detection: based on the edge coordinate equation of the protective cover obtained from the single pixel edge image, the distance and angle between the edges of each protective plate of the protective cover are determined to determine whether the guide rail steel plate protective cover is in normal operating condition; The S5 edge detection includes the following steps: For a single-pixel edge image, the edge One end of the fixed guide rail steel plate protective cover is used as the edge of the fixed end of the guide rail steel plate protective cover. As Establish a coordinate system along the axes to obtain the edges between each piece of the guide rail steel plate protective cover. … The coordinate equations of the coordinate system, where The number of protective plates for the guide rail steel plate protective cover; A) Obtaining the edge … Each with a fixed edge The angle between ;like If this occurs, it is determined that the guide rail steel plate protective cover is abnormal, and an alarm is triggered; among which... Allowable error angle; otherwise, normal operation. B) Calculate the edges separately … Distance between two adjacent edges … And determine whether the following formula is satisfied: ； If the conditions are met, it is determined that the guide rail steel plate protective cover is malfunctioning and an alarm is triggered; otherwise, normal operation continues. Allowable error distance.

2. The edge detection method for a directional guide rail steel plate protective cover according to claim 1, characterized in that, The improved LBP algorithm and the ResNet50 network form the backbone network.

3. The edge detection method for a face-guided guide rail steel plate protective cover according to claim 1, characterized in that, The ResNet50 network was trained using the Annotator-robust loss function: ； ； in, These are hyperparameters used to balance the ratio of positive to negative samples. For the sigmoid function, This represents all the parameters to be learned; This represents the feature vector of the current pixel; The mean of the five annotations in the preprocessing of the S2 dataset was calculated, and then... This indicates that a new marginal probability map is generated, with a range of... 0 indicates that no annotation was performed in any of the 5 instances, and 1 indicates that it was labeled as an edge pixel in all 5 instances. At the same time, the edge probability value exceeds the threshold. As a positive sample Marginal probability values ​​less than As a negative sample .

4. An edge detection device for a face-mounted guide rail steel plate protective cover, the device being used to implement the edge detection method for a face-mounted guide rail steel plate protective cover as described in any one of claims 1-3, characterized in that, include: The multi-dimensional image acquisition module is used to acquire color images of the guide rail steel plate protective cover in the working area through a camera based on the received image acquisition control signal, and output them to the edge detection and analysis module. The edge detection and analysis module includes: The image processing unit is used to acquire edge information from the color image of the guide rail steel plate protective cover through the backbone network; The edge image processing unit performs non-maximum suppression processing on the obtained edge information to obtain a single-pixel edge image; The edge image analysis unit is used to determine whether the protective cover is functioning properly based on the edge information in the single-pixel edge image.

Images