Dynamic division method and system for multi-stage early warning take-up area based on visual segmentation

Abstract

The invention discloses a dynamic division method and system for a multi-stage early warning take-up area based on visual segmentation, belongs to the technical field of machine vision, and aims to solve the technical problems that a traditional dynamic planning method needs frequent calibration, is weak in anti-interference capability and lacks an intelligent grading predetermined mechanism. Comprising the following steps: adaptively triggering a camera to obtain an image through a space frame taking mode; real-time semantic segmentation is carried out on the current image based on the lightweight segmentation network model, a time sequence mask of the current image is obtained and serves as a current mask, and the time sequence mask is divided into two types including a wire coil and a winding wire; calculating the intersection-union ratio between the current mask and the historical reference mask based on a mask updating strategy of time sequence perception, and triggering the updating of the early warning area model; when the updating condition of the early warning area is met, performing weighted fusion on the current mask and a historical reference mask to generate a final mask; and key point extraction is carried out on the final mask, and a take-up early warning area is constructed in a graded manner.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine vision, in particular to a dynamic division method and system for multi-level early warning winding area based on visual segmentation. BACKGROUND

[0002] In the field of industrial winding safety early warning, traditional mechanical sensors and fixed visual detection methods have three technical bottlenecks: first, the detection system cannot dynamically adapt to the production needs of different wire diameters, second, environmental interference leads to high false alarm rate, and third, the detection system is seriously out of sync with the device running state. In view of these industry pain points, it is urgent to develop a multi-level early warning winding system that can intelligently perceive, dynamically partition, and respond in real time, in order to achieve a technical breakthrough from passive detection to active protection.

[0003] The defects of the traditional dynamic programming method, such as the need for frequent calibration, weak anti-interference ability, and lack of intelligent grading mechanism, are technical problems that need to be solved. SUMMARY

[0004] The technical task of the present application is to solve the technical problems of the traditional dynamic programming method, such as the need for frequent calibration, weak anti-interference ability, and lack of intelligent grading mechanism, by providing a dynamic division method and system for multi-level early warning winding area based on visual segmentation.

[0005] In a first aspect, the present application provides a dynamic division method for multi-level early warning winding area based on visual segmentation, comprising the following steps:

[0006] Frame taking: the camera is triggered to obtain images in a frame-taking manner, and the image format is RGB;

[0007] Mask generation: real-time semantic segmentation of the current image is performed based on a lightweight segmentation network model to obtain a time sequence mask M curr As the current mask, the time sequence mask has two categories, namely wire disc and wound wire;

[0008] Time sequence verification: based on the mask update strategy of time sequence perception, the intersection over union between the current mask and the historical reference mask M hist is calculated, if the intersection over union is greater than or equal to the threshold value, it is determined that the winding form of the wire has not changed significantly, and the early warning area model update condition is triggered, if the intersection over union is less than the threshold value, it is determined that the winding form of the wire is abnormal, and the current early warning area model is maintained, and the historical reference mask M hist is not updated;

[0009] Dynamic fusion: when the early warning area update condition is met, the current mask and the historical reference mask M hist are weighted and fused to generate a final mask M new ;

[0010] Geometric calculation: for the final mask M new Key points are extracted, and hierarchical early warning areas are constructed.

[0011] As a preferred method, frame capture without air contact includes the following operations:

[0012] A high-resolution spool rotary encoder is installed on the shaft of the rotary motor of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is acquired in real time through the pulse signal of the spool rotary encoder. The angular displacement data collected by the rotary encoder is output through a high-speed data interface.

[0013] Instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows:

[0014] v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt,

[0015] Where θ(t) represents angular displacement data, Δt represents the adjustable time window, and R(t) represents the real-time radius of the coil;

[0016] A dual-threshold speed determination mode is set up, namely low-speed mode and high-speed mode. In low-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up at fixed time intervals. In high-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up using a displacement trigger mode.

[0017] As a preferred method, dynamic fusion includes the following operations:

[0018] Based on the difference in cross-union ratio (CUI) between the current mask and the historical baseline mask, ΔIoU, calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU), where ΔIoU ∈ [0.25, 1].

[0019] The current mask and the historical baseline mask are linearly weighted and fused to generate the initial fusion result M. new =α·M curr +(1-α)M hist ;

[0020] For the initial fusion result M new Post-fusion processing is performed, including binarization using an adaptive thresholding method and two-step morphological optimization to obtain the final mask M. new .

[0021] Preferably, geometric calculations include the following operations:

[0022] For the mask M, which is a sub-region of the wound wire new-1 Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1Calculate the minimum bounding rectangle Rect new-1 The slope k1;

[0023] Extracting the minimum bounding rectangle (Rect) new-1 The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1 to obtain the extended boundaries L1 and L2. The endpoints of L1 and L2 are connected to form the first-level warning area.

[0024] Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 ;

[0025] Extracting the minimum bounding rectangle (Rect) new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

[0026] In a second aspect, the present invention provides a dynamic partitioning system for a multi-level early warning receiving area based on visual segmentation, comprising a frame capture module, a mask generation module, a timing verification module, a dynamic fusion module, and a geometric calculation module.

[0027] The air-based frame-capturing module is used to perform the following: adaptively trigger the camera to acquire images using an air-based frame-capturing method, with the image format being RGB;

[0028] The mask generation module performs the following: real-time semantic segmentation of the current image based on a lightweight segmentation network model to obtain the temporal mask M of the current image. curr As the current mask, there are two types of timing masks: coil and wound wire;

[0029] The timing verification module performs the following: a timing-aware mask update strategy, and calculates the current mask and the historical baseline mask M. hist The crossover ratio (CRR) is used to determine the wire winding pattern. If the CRR is greater than or equal to a threshold, it is determined that the wire winding pattern has not changed significantly, triggering the update condition for the warning area model. If the CRR is less than the threshold, it is determined that the wire winding pattern is abnormal, maintaining the current warning area model. The historical baseline mask M is used. hist No updates;

[0030] The dynamic fusion module performs the following: When the early warning area update conditions are met, it merges the current mask with the historical baseline mask M. hist Weighted fusion is used to generate the final mask M. new ;

[0031] The geometry calculation module is used to perform the following: For the final mask M... new Key points are extracted, and hierarchical early warning areas are constructed.

[0032] Preferably, the air-based frame-fetching module is used to perform the following operations:

[0033] A high-resolution spool rotary encoder is installed on the shaft of the rotary motor of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is acquired in real time through the pulse signal of the spool rotary encoder. The angular displacement data collected by the rotary encoder is output through a high-speed data interface.

[0034] Instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows:

[0035] v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt,

[0036] Where θ(t) represents angular displacement data, Δt represents the adjustable time window, and R(t) represents the real-time radius of the coil;

[0037] A dual-threshold speed determination mode is set up, namely low-speed mode and high-speed mode. In low-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up at fixed time intervals. In high-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up using a displacement trigger mode.

[0038] Preferably, the dynamic fusion module is used to perform the following operations:

[0039] Based on the difference in cross-union ratio (CUI) between the current mask and the historical baseline mask, ΔIoU, calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU), where ΔIoU ∈ [0.25, 1].

[0040] The current mask and the historical baseline mask are linearly weighted and fused to generate the initial fusion result M. new =α·M curr +(1-α)M hist ;

[0041] For the initial fusion result M new Post-fusion processing is performed, including binarization using an adaptive thresholding method and two-step morphological optimization to obtain the final mask M. new .

[0042] Preferably, the geometry calculation module is used to perform the following operations:

[0043] For the mask M, which is a sub-region of the wound wire new-1 Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1 Calculate the minimum bounding rectangle Rect new-1 The slope k1;

[0044] Extracting the minimum bounding rectangle (Rect) new-1 The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1 to obtain the extended boundaries L1 and L2. The endpoints of L1 and L2 are connected to form the first-level warning area.

[0045] Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 ;

[0046] Extracting the minimum bounding rectangle (Rect) new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

[0047] The dynamic division method and system for multi-level early warning receiving area based on visual segmentation of the present invention has the following advantages:

[0048] 1. Based on visual segmentation technology, the boundary of the work area is identified and dynamically adjusted in real time, and multi-level early warning and line-receiving areas (such as core areas and buffer zones) are accurately divided to adapt to complex environmental changes and improve the accuracy and flexibility of safety protection.

[0049] 2. A multi-level risk assessment mechanism is used to achieve differentiated responses, reducing false alarms and missed alarms. At the same time, a lightweight model is used to ensure real-time performance and low resource consumption. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] The invention will be further described below with reference to the accompanying drawings.

[0052] Figure 1 This is a flowchart of a dynamic division method for a multi-level early warning receiving area based on visual segmentation, as described in Example 1.

[0053] Figure 2 This is a schematic diagram of the warning area in Example 1, a dynamic division method for multi-level warning line receiving areas based on visual segmentation. Detailed Implementation

[0054] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments are not intended to limit the present invention. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0055] This invention provides a method and system for dynamically dividing a multi-level early warning receiving area based on visual segmentation, which solves the technical problems of traditional dynamic programming methods, such as the need for frequent calibration, weak anti-interference ability, and lack of intelligent hierarchical pre-determining mechanism.

[0056] Example 1:

[0057] This invention provides a dynamic partitioning method for multi-level early warning receiving area based on visual segmentation, comprising five steps: remote frame capture, mask generation, timing verification, dynamic fusion, and geometric calculation.

[0058] Step S100: Remote frame capture: Adaptively trigger the camera to acquire images using a remote frame capture method. The image format is RGB.

[0059] As a specific implementation of air frame capture, this step includes the following operations:

[0060] (1) A high-resolution spool rotary encoder is installed on the side of the rotary motor shaft of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is obtained in real time through the pulse signal of the spool rotary encoder. The encoder is installed with a flexible coupling to compensate for axial / radial deviation and avoid measurement errors caused by mechanical vibration. The angular displacement data collected by the rotary encoder is output through a high-speed data interface (such as EtherCAT) to ensure microsecond-level delay and provide accurate input for subsequent speed calculation.

[0061] (2) The instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows:

[0062] v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt,

[0063] Where θ(t) represents angular displacement data (rad), Δt represents the adjustable time window (typical value 5-50ms, adaptively adjusted), and R(t) represents the real-time dynamic update of the coil radius, taking into account the influence of the number of winding layers;

[0064] (3) Set a dual threshold speed judgment mode, namely low speed mode and high speed mode. In low speed mode, the camera is triggered to acquire the image of the wire take-up industrial scene at a fixed time interval (e.g., 20ms / frame). In high speed mode, the camera is triggered to acquire the image of the wire take-up industrial scene by displacement triggering (Δs = 0.5mm / frame). The camera is triggered once every Δs distance.

[0065] By using a dual-threshold mode (low-speed fixed time interval vs. high-speed displacement trigger), the system automatically matches production line speed changes, avoiding data redundancy at low speeds caused by traditional fixed-frequency sampling, and reducing hardware costs and processing time for model calculations.

[0066] Step S200 Mask Generation: Perform real-time semantic segmentation of the current image based on a lightweight segmentation network model to obtain the temporal mask M of the current image. curr As the current mask, there are two categories of timing masks: coil and wound wire.

[0067] In this embodiment, a lightweight visual segmenter (such as MobileNetV3 or PPLiteSeg) is used to perform real-time semantic segmentation on the current frame of the wire winding industrial scene image. The image segmentation model outputs a mask Mcurr, where: target categories: accurately identifying the metal wall of the wire reel (category 1) and the dynamically wound wire (category 2).

[0068] Step S300: Timing verification: Based on a timing-aware mask update strategy, calculate the current mask and the historical baseline mask M. hist The crossover ratio (CRR) is used to determine the wire winding pattern. If the CRR is greater than or equal to a threshold, it is determined that the wire winding pattern has not changed significantly, triggering the update condition for the warning area model. If the CRR is less than the threshold, it is determined that the wire winding pattern is abnormal, maintaining the current warning area model. The historical baseline mask M is used. hist No updates.

[0069] In this embodiment, a time-aware mask update strategy is adopted in this step. By analyzing the spatial consistency between the current frame segmentation mask and the historical reference mask, intelligent anomaly detection and model update control are achieved.

[0070] Step S400 Dynamic Fusion: When the early warning area update conditions are met, the current mask is compared with the historical reference mask M. hist Weighted fusion is used to generate the final mask M. new .

[0071] As a specific implementation of dynamic fusion, this step includes the following operations:

[0072] (1) Calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU) based on the difference between the current mask and the historical reference mask ΔIoU, where ΔIoU ∈ [0.25, 1].

[0073] Fusion is not triggered when ΔIoU is below 0.25, and the current frame is fully trusted when ΔIoU is above 1.

[0074] (2) Perform linear weighted fusion of the current mask and the historical reference mask to generate the initial fusion result M. new =α·M curr +(1-α)M hist In the weight transition region (such as around α = 0.5), Gaussian kernel convolution is used to eliminate the seam effect. The kernel size is adaptively adjusted with ΔIoU: σ = 2 × (1 - |2σ - 1|).

[0075] (3) The initial fusion result M new Post-processing is performed, including binarization using an adaptive thresholding method, followed by dilation and erosion to eliminate small voids, and two-step morphological optimization. Median filtering is then used to suppress impulse noise, resulting in the final mask M. new .

[0076] Step S500 Geometric Calculation: For the final mask M new Key points are extracted, and hierarchical early warning areas are constructed.

[0077] As a specific implementation of geometric calculation, this step includes the following operations:

[0078] (1) The mask M, which is a sub-region of the winding wire new-1 Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1 Calculate the minimum bounding rectangle Rect new-1 The slope k1, such as Figure 2 Points p1 and p2 shown are Rect new-1 The lower left and lower right vertices;

[0079] (2) Extract the minimum bounding rectangle Rect new-1The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1, resulting in the extended boundaries L1 and L2. Connecting the endpoints of L1 and L2 forms the first-level warning area, where d1 = 0.4 * Rect new-1 Short side length;

[0080] (3) Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 At the same time, the deviation between the rectangle angle and k2 is constrained to be less than or equal to 5 degrees;

[0081] (4) Extract the minimum bounding rectangle Rect new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

[0082] The method in this embodiment achieves precise synchronization between image acquisition and device movement through air-to-air frame-taking synchronization technology, and uses a dynamic mask fusion algorithm to enable the detection area to automatically evolve with the line retraction state.

[0083] Example 2:

[0084] The present invention provides a dynamic partitioning system for a multi-level early warning receiving area based on visual segmentation, comprising a frame capture module, a mask generation module, a timing verification module, a dynamic fusion module, and a geometric calculation module.

[0085] The air-based frame capture module is used to perform the following: adaptively trigger the camera to acquire images using an air-based frame capture method, with the image format being RGB.

[0086] As a specific implementation of the air-based frame capture module, this module is used to perform the following operations:

[0087] (1) A high-resolution spool rotary encoder is installed on the side of the rotary motor shaft of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is obtained in real time through the pulse signal of the spool rotary encoder. The encoder is installed with a flexible coupling to compensate for axial / radial deviation and avoid measurement errors caused by mechanical vibration. The angular displacement data collected by the rotary encoder is output through a high-speed data interface (such as EtherCAT) to ensure microsecond-level delay and provide accurate input for subsequent speed calculation.

[0088] (2) The instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows:

[0089] v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt,

[0090] Where θ(t) represents angular displacement data (rad), Δt represents the adjustable time window (typical value 5-50ms, adaptively adjusted), and R(t) represents the real-time dynamic update of the coil radius, taking into account the influence of the number of winding layers;

[0091] (3) Set a dual threshold speed judgment mode, namely low speed mode and high speed mode. In low speed mode, the camera is triggered to acquire the image of the wire take-up industrial scene at a fixed time interval (e.g., 20ms / frame). In high speed mode, the camera is triggered to acquire the image of the wire take-up industrial scene by displacement triggering (Δs = 0.5mm / frame). The camera is triggered once every Δs distance.

[0092] By using a dual-threshold mode (low-speed fixed time interval vs. high-speed displacement trigger), the system automatically matches production line speed changes, avoiding data redundancy at low speeds caused by traditional fixed-frequency sampling, and reducing hardware costs and processing time for model calculations.

[0093] The mask generation module performs the following: real-time semantic segmentation of the current image based on a lightweight segmentation network model to obtain the temporal mask M of the current image. curr As the current mask, there are two categories of timing masks: coil and wound wire.

[0094] In this embodiment, the module uses a lightweight visual segmenter (such as MobileNetV3 or PPLiteSeg) to perform real-time semantic segmentation on the current frame of the wire winding industrial scene image. The image segmentation model outputs a mask Mcurr, where: target categories: accurately identifying the metal wall of the wire reel (category 1) and the dynamically wound wire (category 2).

[0095] The timing verification module performs the following: a timing-aware mask update strategy, and calculates the current mask and the historical baseline mask M. histThe crossover ratio (CRR) is used to determine the wire winding pattern. If the CRR is greater than or equal to a threshold, it is determined that the wire winding pattern has not changed significantly, triggering the update condition for the warning area model. If the CRR is less than the threshold, it is determined that the wire winding pattern is abnormal, maintaining the current warning area model. The historical baseline mask M is used. hist No updates.

[0096] In this embodiment, the module adopts a time-aware mask update strategy. By analyzing the spatial consistency between the current frame segmentation mask and the historical reference mask, it achieves intelligent anomaly detection and model update control.

[0097] The dynamic fusion module performs the following: When the early warning area update conditions are met, it merges the current mask with the historical baseline mask M. hist Weighted fusion is used to generate the final mask M. new .

[0098] As a specific implementation of the dynamic fusion module, this module is used to perform the following operations:

[0099] (1) Calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU) based on the difference between the current mask and the historical reference mask ΔIoU, where ΔIoU ∈ [0.25, 1].

[0100] Fusion is not triggered when ΔIoU is below 0.25, and the current frame is fully trusted when ΔIoU is above 1.

[0101] (2) Perform linear weighted fusion of the current mask and the historical reference mask to generate the initial fusion result M. new =α·M curr +(1-α)M hist In the weight transition region (such as around α = 0.5), Gaussian kernel convolution is used to eliminate the seam effect. The kernel size is adaptively adjusted with ΔIoU: σ = 2 × (1 - |2σ - 1|).

[0102] (3) The initial fusion result M new Post-processing is performed, including binarization using an adaptive thresholding method, followed by dilation and erosion to eliminate small voids, and two-step morphological optimization. Median filtering is then used to suppress impulse noise, resulting in the final mask M. new .

[0103] The geometry calculation module is used to perform the following: For the final mask M... new Key points are extracted, and hierarchical early warning areas are constructed.

[0104] As a specific implementation of the geometry calculation module, this module is used to perform the following operations:

[0105] (1) The mask M, which is a sub-region of the winding wire new-1Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1 Calculate the minimum bounding rectangle Rect new-1 The slope k1, such as Figure 2 Points p1 and p2 shown are Rect new-1 The lower left and lower right vertices;

[0106] (2) Extracting the minimum bounding rectangle Rect new-1 The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1, resulting in the extended boundaries L1 and L2. Connecting the endpoints of L1 and L2 forms the first-level warning area, where d1 = 0.4 * Rect new-1 Short side length;

[0107] (3) Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 At the same time, the deviation between the rectangle angle and k2 is constrained to be less than or equal to 5 degrees;

[0108] (4) Extract the minimum bounding rectangle Rect new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

[0109] The system in this embodiment can execute the method disclosed in Embodiment 1 to achieve dynamic division of multi-level early warning areas.

[0110] The above provides a detailed description of the dynamic division method and system for multi-level early warning receiving area based on visual segmentation provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A dynamic division method for multi-level early warning receiving area based on visual segmentation, characterized in that, Includes the following steps: Air-based frame capture: Adaptively triggers the camera to acquire images using an air-based frame capture method; the image format is RGB. Mask generation: Real-time semantic segmentation of the current image is performed based on a lightweight segmentation network model to obtain the temporal mask M of the current image. curr As the current mask, there are two types of timing masks: coil and wound wire; Time-series verification: a time-aware mask update strategy, calculating the current mask and the historical baseline mask M. hist The crossover ratio (CRR) is used to determine the wire winding pattern. If the CRR is greater than or equal to a threshold, it is determined that the wire winding pattern has not changed significantly, triggering the update condition for the warning area model. If the CRR is less than the threshold, it is determined that the wire winding pattern is abnormal, maintaining the current warning area model. The historical baseline mask M is used. hist No updates; Dynamic fusion: When the conditions for updating the early warning area are met, the current mask is compared with the historical baseline mask M. hist Weighted fusion is used to generate the final mask M. new ; Geometric calculation: for the final mask M new Key points are extracted, and hierarchical early warning areas are constructed.

2. The dynamic division method for multi-level early warning receiving area based on visual segmentation according to claim 1, characterized in that, Air-based frame capture includes the following operations: A high-resolution spool rotary encoder is installed on the shaft of the rotary motor of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is acquired in real time through the pulse signal of the spool rotary encoder. The angular displacement data collected by the rotary encoder is output through a high-speed data interface. Instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows: v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt, Where θ(t) represents angular displacement data, Δt represents the adjustable time window, and R(t) represents the real-time radius of the coil; A dual-threshold speed determination mode is set up, namely low-speed mode and high-speed mode. In low-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up at fixed time intervals. In high-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up using a displacement trigger mode.

3. The dynamic division method for multi-level early warning receiving area based on visual segmentation according to claim 1, characterized in that, Dynamic fusion includes the following operations: Based on the difference in cross-union ratio (CUI) between the current mask and the historical baseline mask, ΔIoU, calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU), where ΔIoU ∈ [0.25, 1]. The current mask and the historical baseline mask are linearly weighted and fused to generate the initial fusion result M. new =α·M curr +(1-α)M hist ; For the initial fusion result M new Post-fusion processing is performed, including binarization using an adaptive thresholding method and two-step morphological optimization to obtain the final mask M. new .

4. The dynamic division method for multi-level early warning receiving area based on visual segmentation according to claim 1, characterized in that, Geometric calculations include the following operations: For the mask M, which is a sub-region of the wound wire new-1 Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1 Calculate the minimum bounding rectangle Rect new-1 The slope k1; Extracting the minimum bounding rectangle (Rect) new-1 The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1 to obtain the extended boundaries L1 and L2. The endpoints of L1 and L2 are connected to form the first-level warning area. Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 ; Extracting the minimum bounding rectangle (Rect) new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

5. A dynamic segmentation system for multi-level early warning receiving areas based on visual segmentation, characterized in that, It includes a frame capture module, a mask generation module, a timing verification module, a dynamic fusion module, and a geometric calculation module; The air-based frame-capturing module is used to perform the following: adaptively trigger the camera to acquire images using an air-based frame-capturing method, with the image format being RGB; The mask generation module performs the following: real-time semantic segmentation of the current image based on a lightweight segmentation network model to obtain the temporal mask M of the current image. curr As the current mask, there are two types of timing masks: coil and wound wire; The timing verification module performs the following: a timing-aware mask update strategy, and calculates the current mask and the historical baseline mask M. hist The crossover ratio (CRR) is used to determine the wire winding pattern. If the CRR is greater than or equal to a threshold, it is determined that the wire winding pattern has not changed significantly, triggering the update condition for the warning area model. If the CRR is less than the threshold, it is determined that the wire winding pattern is abnormal, maintaining the current warning area model. The historical baseline mask M is used. hist No updates; The dynamic fusion module performs the following: When the early warning area update conditions are met, it merges the current mask with the historical baseline mask M. hist Weighted fusion is used to generate the final mask M. new ; The geometry calculation module is used to perform the following: For the final mask M... new Key points are extracted, and hierarchical early warning areas are constructed.

6. The dynamic division system for multi-level early warning receiving area based on visual segmentation according to claim 5, characterized in that, The air frame capture module is used to perform the following operations: A high-resolution spool rotary encoder is installed on the shaft of the rotary motor of the industrial take-up machine. The spool rotary encoder rotates synchronously with the motor. The angular displacement data of the spool is acquired in real time through the pulse signal of the spool rotary encoder. The angular displacement data collected by the rotary encoder is output through a high-speed data interface. Instantaneous linear velocity is calculated based on angular displacement data using a sliding window difference algorithm. The calculation formula is as follows: v(t)=(θ(t)-θ(t-Δt)×R(t)) / Δt, Where θ(t) represents angular displacement data, Δt represents the adjustable time window, and R(t) represents the real-time radius of the coil; A dual-threshold speed determination mode is set up, namely low-speed mode and high-speed mode. In low-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up at fixed time intervals. In high-speed mode, the camera is triggered to acquire images of the industrial scene of wire take-up using a displacement trigger mode.

7. The dynamic division system for multi-level early warning receiving area based on visual segmentation according to claim 5, characterized in that, The dynamic fusion module is used to perform the following operations: Based on the difference in cross-union ratio (CUI) between the current mask and the historical baseline mask, ΔIoU, calculate the dynamic weighting coefficient α = min(1, 0.3, 0.7 * ΔIoU), where ΔIoU ∈ [0.25, 1]. The current mask and the historical baseline mask are linearly weighted and fused to generate the initial fusion result M. new =α·M curr +(1-α)M hist ; For the initial fusion result M new Post-fusion processing is performed, including binarization using an adaptive thresholding method and two-step morphological optimization to obtain the final mask M. new .

8. The dynamic division system for multi-level early warning receiving area based on visual segmentation according to claim 5, characterized in that, The geometry calculation module is used to perform the following operations: For the mask M, which is a sub-region of the wound wire new-1 Perform a rotation and fit the minimum bounding rectangle to obtain the minimum bounding rectangle Rect. new-1 Calculate the minimum bounding rectangle Rect new-1 The slope k1; Extracting the minimum bounding rectangle (Rect) new-1 The center points p3 and p4 of the upper and lower interfaces are used as the reference key points of the first-level warning area. The axis formed by the reference key points p3 and p4 is used as the center line to extend vertically to both sides by a distance d1 to obtain the extended boundaries L1 and L2. The endpoints of L1 and L2 are connected to form the first-level warning area. Using the orthogonal slope k2 = -1 / (k1 ± 0.01) as a constraint, the mask M of the wire disk region, which is the wire disk region, is... new-2 Perform a rotational minimum bounding rectangle fitting to obtain the minimum bounding rectangle Rect. new-2 ; Extracting the minimum bounding rectangle (Rect) new-1 and minimum bounding rectangle Rect new-2 Based on the four sides, calculate all possible intersection points using line segment intersection detection, and filter those located within the minimum bounding rectangle Rect. new-1 and minimum bounding rectangle Rect new-2 The two intersection points p5 and p6 are connected to the minimum bounding rectangle Rect. new-1 The far vertices p1 and p2 are combined, and a secondary warning area of ​​a convex polygon is constructed based on the direction of the line convergence.

Images