A method, apparatus, medium and device for monitoring a barrier fence

Abstract

The application discloses an isolation fence monitoring method, device, medium and equipment, and belongs to the technical field of computer vision and image processing; the method uses a DINOv2 lightweight pre-training model as a backbone network and YOLOe as a teacher network to complete accurate semantic segmentation of a fence and surrounding objects, extracts target object semantic information through the DINOv2, compares fence semantic information at different times by using a dense matching method, and determines whether the fence is damaged according to a semantic change threshold; meanwhile, the contact condition of the semantic regions of the fence and the surrounding objects is detected, potential damage risks are determined and warnings are triggered in combination with the semantic change of the contact regions, and finally, a monitoring result is generated and real-time alarms are given; the application further provides a monitoring device, a computer readable storage medium and an electronic device adapted to the method; the application can be widely applied to fields such as construction sites and public safety monitoring, and helps to timely find fence damage or potential damage risks, and ensure the safety of construction sites.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and image processing technology, and particularly relates to a method, device, medium and equipment for monitoring isolation fences. Background Technology

[0002] Existing semantic segmentation algorithms extract semantic information from all objects in an image. However, since fences are not the only targets in an image, extracting semantic information from all objects consumes significant computational resources, leading to inefficiency and failing to meet the needs of real-time monitoring. The YOLOe network, as a teacher network, can effectively outline and locate objects in an image, but it lacks precise semantic segmentation capabilities. Therefore, when fences overlap or contact other objects (such as construction machinery, people, and vehicles), YOLOe may fail to accurately segment the fence boundaries, affecting the accurate determination of the fence's state. Existing methods perform poorly under varying lighting conditions or inclement weather. For example, on cloudy days, at night, or in strong sunlight, image quality may be poor, causing detection algorithms to fail to accurately identify the fence's state. Changes in lighting can cause shadows or reflections on the fence in the image, affecting the accuracy of damage detection. Construction information or markings may be affixed to fences at construction sites; these objects may overlap with or obscure parts of the fence surface, easily leading to misjudgments by existing methods as damage or breakage, thus affecting the accuracy of fence damage detection.

[0003] Therefore, there is an urgent need for an algorithm that combines object detection and semantic segmentation, which can not only improve the efficiency of fence detection and optimize computing resources, but also accurately identify fence damage and potential damage signs in situations such as complex backgrounds, changes in lighting, and misjudged object occlusion, so as to ensure the safety monitoring of construction sites. Summary of the Invention

[0004] The purpose of this invention is to provide a method, device, medium, and equipment for monitoring isolation fences, addressing the problems of high computational resource consumption, inaccurate object segmentation, and the influence of lighting changes on the accuracy of damage detection, as well as misjudgments and false identifications, in existing technologies. This invention achieves efficient fence status monitoring by combining DINOv2 semantic segmentation and YOLOe object detection, accurately determining whether the fence is damaged or shows signs of damage. By employing a dense matching method, the semantic information of the fence is compared at different times, effectively reducing computational resource consumption, improving the accuracy of fence damage detection, and ensuring the safety of the construction site.

[0005] To achieve the above objectives, the present invention provides a method for monitoring isolation fences, comprising the following steps: S1. Perform fence state monitoring and perform accurate semantic segmentation of objects based on the DINOv2 lightweight pre-trained model and YOLOe; S2. Based on the DINOv2 backbone network, the semantic information of the objects segmented from the image in S1 using YOLOe as the supervision signal is globally extracted. S3. Input an image into the DINOv2 lightweight pre-trained model and extract the semantic information of objects in the image; S4. Use a dense matching method to match the semantic information of the fence at different times. If the semantics of the fence changes and the amount of change exceeds the threshold, it is determined that the fence has been damaged. S5. When the semantic region of the fence comes into contact with the semantic region of an object, it is determined that the object is approaching the fence and has the potential risk of damaging the fence, and a warning is issued through the warning device. S6. Based on the status of the fence and its relationship with surrounding objects, generate monitoring results and trigger alarms in real time to detect fence damage and signs of damage.

[0006] Preferably, the specific content of S1 is as follows: The DINOv2 lightweight pre-trained model is used as the backbone network, while YOLOe is used as the teacher network. The YOLOe object detection algorithm is used to outline the region of the object. The semantic segmentation head network is obtained by training DINOv2 to perform accurate semantic segmentation of the object.

[0007] Preferably, the dense matching method in S4 is used to compare the state of the fence at different times based on the semantic information of the fence. If the change in the semantic information of the fence exceeds a preset threshold, it is determined that the fence is damaged.

[0008] Preferably, in S5, the potential risk of damaging the fence is determined by performing semantic contact detection with objects near the fence. If the semantic information of the fence in the contact area changes, it is determined that the fence is at potential risk of damage.

[0009] Preferably, in S5, warnings are issued via loudspeakers and warning devices to promptly notify on-site personnel to take necessary measures.

[0010] Preferably, a fence monitoring device includes: an image processing device and a central processing unit; the image processing device includes an image acquisition module and an output module; the central processing unit includes an anomaly detection module.

[0011] Preferably, the image acquisition module includes a camera for acquiring raw images.

[0012] Preferably, the output module includes a display and a transmitter for displaying and transmitting processed images and monitoring results.

[0013] Preferably, a computer-readable storage medium stores a computer program that, when executed by a processor, implements the fence monitoring method described in any of the preceding claims.

[0014] Preferably, an electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the fence monitoring method described in any of the above claims.

[0015] Therefore, the beneficial effects of the above-mentioned method, device, medium, and equipment for monitoring isolation fences are as follows: (1) By combining object detection and semantic segmentation, the efficiency of fence monitoring is improved, the consumption of computing resources is reduced, and real-time performance is ensured; (2) Through the three core designs of DINOv2+YOLOe semantic segmentation technology, fine comparison of dense matching, and early warning of semantic contact, it can accurately detect the damage or potential damage to the fence, especially the damage that is difficult to detect. (3) The detection logic based on semantic features replaces the traditional pixel / contour detection. Combined with high-precision semantic segmentation, fine comparison of dense matching, and targeted semantic extraction, it avoids the interference of complex environment on visual detection from the root. It can accurately detect in complex environment (such as changes in lighting, object occlusion, etc.), avoid misjudgment and missed judgment, and improve the accuracy of monitoring. (4) By combining DINOv2 and YOLOv11, there is no need to retrain the backbone network. Only the semantic segmentation head model needs to be trained. It is compatible with existing object detection and semantic segmentation algorithms and is easy to integrate into existing monitoring systems. (5) Supports real-time processing and is suitable for embedded devices (such as construction site monitoring equipment, drones, etc.) to ensure efficient on-site safety monitoring.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for monitoring isolation fences according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0019] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0020] The following is in conjunction with the appendix Figure 1 The embodiments of the present invention will be described in detail below.

[0021] Example 1 This embodiment provides a method for monitoring isolation fences, including the following steps: S1. Perform fence status monitoring and accurately segment objects such as fences, construction machinery, people, and vehicles based on the DINOv2 lightweight pre-trained model and YOLOe. The DINOv2 lightweight pre-trained model is used as the backbone network, and YOLOe is used as the teacher network. The YOLOe object detection algorithm is used to outline the regions of objects such as fences, construction machinery, people, and vehicles. The semantic segmentation head network is obtained by training DINOv2 to perform accurate semantic segmentation of objects such as fences, construction machinery, people, and vehicles.

[0022] S2. Based on the DINOv2 backbone network, the semantic information of objects such as fences, construction machinery, people and vehicles in the image segmented by YOLOe as the supervision signal in S1 is globally extracted. S3. Input an image into the DINOv2 lightweight pre-trained model to extract semantic information of objects such as fences, construction machinery, people, and vehicles in the image. This avoids unnecessary calculations caused by global semantic extraction, reduces the consumption of computing resources, and facilitates deployment on edge devices. S4. Use a dense matching method to match the semantic information of the fence at different times. If the semantics of the fence changes and the amount of change exceeds the threshold, it is determined that the fence has been damaged. The dense matching method is used to compare the state of the fence at different times based on the semantic information of the fence. If the change in the semantic information of the fence exceeds a preset threshold, it is determined that the fence is damaged.

[0023] S5. When the semantic area of ​​the fence comes into contact with the semantics of people, vehicles and construction machinery, it is determined that these objects are approaching the fence and pose a potential risk of damaging the fence, and a warning is issued through the warning device. The potential risk of fence damage is assessed by performing semantic contact detection with objects such as fences, construction machinery, people, and vehicles in the vicinity of the fence. If the semantic information of the fence in the contact area changes, it is determined that the fence is at potential risk of damage.

[0024] Warnings are issued via loudspeakers and warning devices to promptly notify on-site personnel to take necessary measures to prevent further damage to the fence.

[0025] S6. Based on the status of the fence and its relationship with surrounding fences, construction machinery, people, vehicles, and other objects, generate monitoring results and trigger alarms in real time to detect fence damage and signs of damage.

[0026] The present invention provides a monitoring device for isolation fences, comprising: an image processing device and a central processing unit; the image processing device includes an image acquisition module and an output module; the central processing unit includes an anomaly detection module.

[0027] The image acquisition module includes a camera for acquiring raw images.

[0028] The output module includes a display and a transmitter for displaying and transmitting processed images and monitoring results.

[0029] The present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the isolation fence monitoring method described in any of the above claims.

[0030] The present invention provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the isolation fence monitoring method described in any of the above claims.

[0031] Example 2 Taking construction site fence monitoring as an example: Step 1: Fence status monitoring and object segmentation; 1) Input image: Obtain the original image of the construction site.

[0032] 2) Object Detection and Segmentation: The YOLOe object detection algorithm is used to outline the regions of objects such as fences, construction machinery, people, and vehicles. Then, DINOv2 is used as the backbone network to perform accurate semantic segmentation of objects such as fences, construction machinery, people, and vehicles.

[0033] YOLOe can accurately detect target objects and define regions, while DINOv2 has excellent semantic extraction capabilities. Combining the two can effectively segment and extract the semantic information of the fence while reducing the computational burden.

[0034] 3) Semantic extraction: The DINOv2 network extracts semantic information of objects such as fences, construction machinery, people, and vehicles in the image, avoiding interference from irrelevant objects and focusing on extracting fence-related information.

[0035] Step 2: Dense matching and fence destruction detection; 1) Dense Matching: This method performs dense matching on the semantic information of the fence at different times, comparing changes in the fence. It calculates the semantic differences of the fence at different time points; if the semantic changes exceed a preset threshold, the fence is considered damaged. Dense matching can effectively capture subtle changes in the fence at different time points, improving the accuracy of damage detection.

[0036] 2) Damage indication judgment: When the semantic area of ​​the fence is in contact with or close to surrounding objects (such as construction machinery, people, vehicles, etc.), it is determined that these objects may cause damage to the fence, and there is a risk of damage if the distance is too close.

[0037] The specific implementation process is as follows: 1) Cameras are installed at the construction site to capture raw images of the fence and its surroundings.

[0038] 2) The algorithm uses YOLOe to outline the regions of objects such as fences, construction machinery, people, and vehicles, and uses DINOv2 to extract semantic information of fences and other targets.

[0039] 3) By using a dense matching method, the semantic information of the fence at different time points is compared to detect whether the fence has changed. If the change exceeds the threshold, the fence is determined to be damaged.

[0040] 4) If the fenced area comes into contact with or gets close to construction machinery or personnel, the system will issue a real-time warning through warning devices (such as horns).

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for monitoring isolation fences, characterized in that, Includes the following steps: S1. Perform fence state monitoring and perform accurate semantic segmentation of objects based on the DINOv2 lightweight pre-trained model and YOLOe; S2. Based on the DINOv2 backbone network, the semantic information of the objects segmented from the image in S1 using YOLOe as the supervision signal is globally extracted. S3. Input an image into the DINOv2 lightweight pre-trained model and extract the semantic information of objects in the image; S4. Use a dense matching method to match the semantic information of the fence at different times. If the semantics of the fence changes and the amount of change exceeds the threshold, it is determined that the fence has been damaged. S5. When the semantic region of the fence comes into contact with the semantic region of an object, it is determined that the object is approaching the fence and has the potential risk of damaging the fence, and a warning is issued through the warning device. S6. Based on the status of the fence and its relationship with surrounding objects, generate monitoring results and trigger alarms in real time to detect fence damage and signs of damage.

2. The method for monitoring isolation fences according to claim 1, characterized in that, The specific content of S1 is as follows: The DINOv2 lightweight pre-trained model is used as the backbone network, while YOLOe is used as the teacher network. The YOLOe object detection algorithm is used to outline the region of the object. The semantic segmentation head network is obtained by training DINOv2 to perform accurate semantic segmentation of the object.

3. The method for monitoring isolation fences according to claim 2, characterized in that: The dense matching method in S4 is used to compare the fence state at different times based on the fence's semantic information. If the change in the fence's semantic information exceeds a preset threshold, it is determined that the fence is damaged.

4. The method for monitoring isolation fences according to claim 3, characterized in that: In S5, the potential risk of damaging the fence is determined by performing semantic contact detection with objects near the fence. If the semantic information of the fence in the contact area changes, the fence is determined to be at potential risk of damage.

5. The method for monitoring isolation fences according to claim 4, characterized in that: In S5, warnings are issued via loudspeakers and warning devices to promptly notify on-site personnel to take necessary measures.

6. A fence monitoring device, employing a fence monitoring method as described in any one of claims 1-5, characterized in that, include: Image processing device and central processing unit; the image processing device includes an image acquisition module and an output module; The central processing unit includes an anomaly detection module.

7. The isolation fence monitoring device according to claim 6, characterized in that: The image acquisition module includes a camera for acquiring raw images.

8. The isolation fence monitoring device according to claim 6, characterized in that: The output module includes a display and a transmitter for displaying and transmitting processed images and monitoring results.

9. A computer-readable storage medium, characterized in that: The device contains a computer program that, when executed by a processor, implements a fence monitoring method according to any one of claims 1-5.

10. An electronic device, characterized in that: It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the isolation fence monitoring method according to any one of claims 1-5.

Images