Intelligent monitoring and early warning method and system for high-altitude operation

Abstract

The invention discloses an intelligent monitoring and early warning method and system for high-altitude operation, and relates to the technical field of safety production monitoring and target detection.The method comprises the steps that a climbing preparation area, a climbing area and a high-altitude walking area are arranged, and target detection is conducted on the three areas; if a person is detected in the climbing preparation area and does not wear safety equipment, whether the person enters the climbing area is judged, and if the person enters the climbing area, early warning information is generated; if the person is detected in the climbing area and does not wear the safety equipment, early warning information is generated and filed; if the person is detected in the high-altitude walking area and does not wear the safety equipment, early warning information is generated and filed, and the real-time performance and accuracy of early warning of the illegal high-altitude operation event are improved. According to the target detection network, the fusion down-sampling module and the self-attention pyramid pooling module are introduced, so that the detection capability of small-size targets such as safety equipment is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of production safety monitoring and target detection, and in particular to an intelligent monitoring and early warning method and system for high-altitude operations. Background Art

[0002] Safety monitoring and violation prevention are crucial when working at heights at construction sites. Traditional monitoring systems mostly rely on manual surveillance or basic video analysis techniques, such as setting up fixed surveillance cameras and using basic image processing techniques to identify the wearing of seat belts. However, these methods are usually unable to automatically identify complex violations, have low recognition rates for small-sized targets, and are easily affected by environmental factors such as lighting and occlusion.

[0003] In addition, some solutions using sensor technology, such as patent document US20190236455A1, use wearable devices to detect whether workers are using safety equipment correctly. Although this improves the accuracy of monitoring, the deployment cost of such systems is high and requires additional hardware support, which limits their application scope in large construction sites.

[0004] With the development of deep learning technology, the capabilities of image recognition and target detection have been significantly improved. For example, the YOLO series of algorithms have shown extremely high efficiency and accuracy in real-time target detection. However, although these algorithms perform well in processing large-sized targets, there are still challenges in identifying small-sized or partially occluded targets such as seat belt buckles and fall arresters.

[0005] In summary, the existing technology has the following main shortcomings in the safety monitoring of high-altitude operations at construction sites:

[0006] 1. Low recognition rate and environmental dependence: Traditional surveillance systems rely on basic video analysis technology, which usually cannot effectively identify small or partially obscured targets, such as seat belt buckles and anti-fall equipment. In addition, their recognition performance is greatly affected by environmental factors such as lighting, weather, and viewing angle.

[0007] 2. Limitations of manual monitoring: Systems that rely on manual monitoring are not only inefficient, but also susceptible to factors such as monitoring staff fatigue and distraction, resulting in unstable monitoring quality.

[0008] 3. High cost and scalability issues: Although some advanced monitoring systems, such as wearable sensor systems, can improve the accuracy of monitoring, these systems often involve expensive hardware equipment and complex deployment procedures, which limits their widespread application in large-scale construction sites.

[0009] 4. Lack of real-time processing and response capabilities: Most current systems lack the ability to process real-time data and automatically warn, which may lead to delayed processing of important information in emergency situations. Summary of the invention

[0010] In order to overcome the defects in the above-mentioned prior art, the present invention provides an intelligent monitoring and early warning method for aerial operations, which can achieve higher-precision real-time monitoring and automatic violation identification in complex construction environments, and significantly improve the efficiency and accuracy of construction safety management.

[0011] To achieve the above object, the present invention adopts the following technical solutions, including:

[0012] An intelligent monitoring and early warning method for high-altitude operations, which sets a climbing preparation area, a climbing area and a high-altitude walking area, and performs target detection in the three areas respectively; the targets include personnel and safety equipment;

[0013] If a person is detected in the climbing preparation area, it is determined whether the person is wearing safety equipment. If the person is wearing safety equipment, the climbing preparation area is continued to be detected. If the person is not wearing safety equipment, it is determined whether the person has entered the climbing area. If the person has entered the climbing area, an early warning message is generated and archived. If the person has not entered the climbing area, the climbing preparation area is continued to be detected.

[0014] If a person is detected in the climbing area, it is determined whether the person is wearing safety equipment: if the person is wearing safety equipment, the target detection in the climbing area will continue; if the person is not wearing safety equipment, an early warning message will be generated and archived;

[0015] If a person is detected in the high-altitude walking area, it is determined whether the person is wearing safety equipment: if the person is wearing safety equipment, target detection of the high-altitude walking area will continue; if the person is not wearing safety equipment, a warning message will be generated and archived.

[0016] Preferably, the safety equipment is two fall arresters;

[0017] If a person is detected in the high-altitude walking area, determine whether the person is wearing a fall arrester: if the person is wearing two fall arresters, continue to perform target detection in the high-altitude walking area; if the person is not detected to be wearing two fall arresters, pause for t seconds and perform fall arrester detection again. If the person is still not detected to be wearing two fall arresters after t seconds, generate an early warning message and archive it. If the person is detected to be wearing two fall arresters after t seconds, continue to perform personnel detection and corresponding fall arrester detection in the high-altitude walking area.

[0018] Preferably, a target detection network is used for target detection;

[0019] At present, the detection network consists of three parts: backbone feature extraction network, PANet multi-scale feature fusion network, and feature decoder;

[0020] The backbone feature extraction network is composed of a fusion downsampling module and a C2F module connected in sequence; the self-attention pyramid pooling module, namely SASPP, is introduced after the backbone feature extraction network. SASPP combines the channel attention mechanism and the spatial attention mechanism. Features of different scales are processed by SASPP and then input into the PANet multi-scale feature fusion module for enhancement; the features of different scales enhanced by the PANet multi-scale feature fusion module are respectively input into the corresponding feature decoders for decoding to identify the target.

[0021] Preferably, the fused downsampling module first uses the convolution layer, the maximum pooling layer and the average pooling layer in parallel to extract features from the input respectively to obtain three downsampling features, and then convolves the three downsampling features through the convolution layer respectively and then fuses them to obtain the fused downsampling features.

[0022] Preferably, the backbone network includes N fusion downsampling modules and N C2F modules, and the fusion downsampling modules and the corresponding C2F modules are alternately connected in sequence.

[0023] Preferably, the self-attention pyramid pooling module combines the channel attention mechanism and the spatial attention mechanism, and specifically includes a CBA module consisting of a convolutional layer, a normalization layer and an activation function.

[0024] Preferably, the feature decoder adopts a decoupling strategy to process the classification task and the positioning task respectively. After being processed by a CBA module, the input feature vector is divided into two branches: one branch is used to predict the confidence of the target existence and the geometric parameters of the detection box, and the other branch is used to predict the category probability of the target.

[0025] The present invention also provides an intelligent monitoring and early warning system for aerial work, which is applicable to the above-mentioned intelligent monitoring and early warning method for aerial work. The system includes: a video acquisition unit, a target detection unit, a processing unit and an early warning unit;

[0026] The video acquisition unit is used to collect video images of the climbing preparation area, the climbing area and the high-altitude walking area in real time, and send the video images to the target detection unit;

[0027] The target detection unit is used to detect in real time whether a target, including personnel and safety equipment, appears in the video image, and send the detection result to the processing unit;

[0028] The processing unit is used to make a real-time judgment based on the detection result to determine whether there is a violation, and send the judgment result to the early warning unit;

[0029] The processing unit is used to issue an early warning in real time according to the judgment result.

[0030] The present invention also provides a computer program product, which includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the above-mentioned intelligent monitoring and early warning method for high-altitude operations is implemented.

[0031] The present invention also provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned intelligent monitoring and early warning method for aerial work is implemented.

[0032] The advantages of the present invention are:

[0033] (1) The present invention reduces reliance on manual monitoring through an automated monitoring method, improves monitoring efficiency and quality, and ensures continuous and uninterrupted safety monitoring of climbing points.

[0034] (2) The present invention designs a cost-effective and easy-to-deploy intelligent monitoring system with real-time data processing and automatic early warning functions, which can immediately issue warnings when violations or dangerous behaviors are detected to prevent accidents.

[0035] (3) The present invention proposes an intelligent monitoring and early warning method for high-altitude operations at construction sites based on an improved deep learning network, which effectively enhances the target detection network's ability to recognize small-sized and partially occluded targets. The present invention can achieve higher-precision real-time monitoring and automatic violation identification in complex construction environments, significantly improving the efficiency and accuracy of construction safety management.

[0036] (4) The present invention develops a new image processing algorithm that can effectively identify people and small-sized targets such as seat belt buckles, fall arresters, etc., and has strong adaptability to environmental changes.

[0037] (5) By integrating the downsampling mechanism, self-attention mechanism and multi-scale pooling strategy, the detection capability of small-sized targets is significantly improved. This technology not only enhances the details of feature extraction, but also optimizes the model's response to small objects in complex backgrounds.

[0038] (6) The classification task and localization task in target detection are decoupled and optimized separately. This design reduces the mutual interference between tasks and improves detection accuracy and reliability.

[0039] (7) Aiming at specific industrial safety scenarios, the present invention designs a set of real-time monitoring and judgment logic, which can identify and respond to the safety status of personnel in real time, issue early warnings in time, and increase operational safety.

[0040] (8) The network structure designed by the present invention can flexibly adapt to different monitoring environments, including places that are visually complex or spatially confined, while ensuring detection performance and adapting to the needs of various industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of illegal climbing determination in the present invention.

[0042] Figure 2 This is a flow chart for determining illegal high-altitude walking in the present invention.

[0043] Figure 3 Schematic diagram of the improved target detection network of the present invention.

[0044] Figure 4 Schematic diagram of the fusion downsampling module.

[0045] Figure 5 Schematic diagram of the self-attention pyramid pooling module.

[0046] Figure 6 Schematic diagram of channel and spatial attention modules. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Example 1

[0049] The present invention relates to an intelligent monitoring and early warning method for aerial work based on deep learning technology, which is particularly suitable for identifying illegal behaviors in aerial work in monitoring videos, such as climbing or walking at high altitude without wearing a fall arrester. The core of the method is that it designs a dedicated target detection network for personnel and safety equipment, as well as finely defined illegal operation judgment rules, to ensure the accuracy and real-time warning of illegal behaviors in aerial work.

[0050] The target detection network proposed in this invention is an improved deep learning network, which is specially used for efficient and accurate monitoring of high-altitude workers and their behaviors. Figure 3 As shown in the figure, the network consists of three parts: backbone network, multi-scale feature fusion network and feature decoder.

[0051] Since the size of human targets and safety equipment targets such as fall arresters is quite different, conventional target detection networks are prone to information loss during feature downsampling. Therefore, the present invention designs a fusion downsampling module, such as Figure 4 As shown, the module uses 3×3 convolution layers, maximum pooling and average pooling layers in parallel to downsample the input respectively, and obtains three downsampled features. The three downsampled features are convolved through 1×1 convolution layers and then fused to obtain fused downsampled features. The fused downsampled module is mainly used to retain the features of small-sized safety equipment, and can also suppress local interference. The fused downsampled module is combined with the C2F module in the YOLOv8 network to form the backbone network of the present invention, and the output of the fused downsampled module is used as the input of the C2F module; wherein the backbone network includes 4 fused downsampled modules and 4 C2F modules, and the fused downsampled modules and the corresponding C2F modules are connected alternately in sequence.

[0052] After the backbone network, a self-attention pyramid pooling module (SASPP) is introduced. This module effectively reduces redundant features and enhances the representation of key local information by combining the channel attention mechanism and the spatial attention mechanism, significantly improving the recognition accuracy of small-sized objects. Figure 5 As shown in Figure 1, the SASPP module includes a CBA module consisting of a convolutional layer (C), a normalization layer (B) and an activation function (A), as well as a channel and spatial attention module (CBAM module). The CBAM module is shown in Figure 6 As shown in Figure 2. In addition, the SASPP module uses pooling kernels of different sizes (5x5, 9x9, 13x13) and keeps the stride as 1, so as not to change the spatial size of the feature map. In this way, the SASPP module is able to capture features at different scales and is further optimized by the CBAM module to highlight key features and suppress irrelevant information.

[0053] Features of different scales are processed by the SASPP module and then input into the PANet multi-scale feature fusion module for enhancement.

[0054] In terms of feature decoding, the present invention adopts a decoupling strategy to process the classification task and the positioning task separately. After being processed by a CBA module, the input feature vector is divided into two branches: one is responsible for predicting the confidence of the target existence and the geometric parameters of the detection box, and the other is dedicated to predicting the class probability of the target. This method improves the independence and accuracy of classification and positioning, ensuring that the personnel and fall arresters in the image can be accurately identified.

[0055] The present invention realizes the monitoring and early warning functions of two kinds of high-altitude operations, namely, climbing and high-altitude walking at the construction site, based on target detection.

[0056] like Figure 1 As shown in the figure, in the climbing monitoring and early warning function, two key monitoring areas are set up through detailed analysis of specific scenarios: the climbing preparation area and the climbing area. The workflow is as follows:

[0057] When a person is detected in the climbing preparation area, it is determined that the person may start climbing work, and the fall arrester detection program is automatically activated. If the person is wearing a fall arrester, or does not enter the climbing area from the climbing preparation area, this is considered normal and reset to the initial monitoring state. If a person enters the climbing area without wearing a fall arrester, the violation is immediately identified and an alarm message is generated to prevent potential safety accidents.

[0058] like Figure 2 As shown in the figure, in the high-altitude walking monitoring and early warning function, the high-altitude walking area is set as the key monitoring area through detailed analysis of specific scenarios. The workflow is as follows:

[0059] When a person is detected in the monitoring screen, first determine whether the current person is in the high-altitude walking area. If the current person is in the high-altitude walking area, turn on the fall arrester detection function. If two fall arresters can be detected, it means that the current person is performing high-altitude walking operations in accordance with regulations; if two fall arresters cannot be detected, it means that the current person may not comply with the regulations. Considering that the position of the fall arrester needs to be manually changed when the person passes the node, the system will suspend detection for 3 seconds. If two fall arresters are detected after 3 seconds, there is no need to alarm. If two fall arresters still cannot be detected after 3 seconds, it means that the current person is very likely to be performing illegal operations, and an alarm message will be generated immediately.

[0060] In this way, the present invention significantly improves the real-time identification and early warning speed of illegal high-altitude operations, effectively prevents high-altitude operations without taking safety measures, and ensures the safety of the workplace.

[0061] The present invention improves the detection accuracy of small-sized safety equipment: By introducing a fused downsampling module and a self-attention pyramid pooling module, the present invention significantly improves the detection capability of small-sized targets such as fall arresters. The network uses a self-attention mechanism to enhance the expression of local information in the feature map, allowing the network to focus on key features more accurately. The multi-scale pooling and feature fusion structure improve the receptive field of the model and enhance the detection performance of small targets. The present invention improves the focus and performance of each task by effectively decoupling the classification task and the positioning task and processing them separately, thereby further improving the accuracy and reliability of fall arrester detection.

[0062] The present invention improves the real-time and accuracy of early warning of illegal high-altitude operations: by designing a specific monitoring logic, the present invention can determine in real time whether the staff has taken appropriate safety measures. The system can automatically identify and respond to different states of personnel, issue alarms in time, and prevent personnel who have not taken safety measures from entering dangerous areas, greatly improving the ability to prevent accidents.

[0063] The present invention can flexibly adapt to a variety of monitoring environments: The network structure and monitoring logic design of the present invention are flexible and can be widely used in various industrial environments, especially in scenarios with space constraints and variable target sizes. This flexibility ensures that the technical solution of the present invention can maintain high performance in different application scenarios and meet a wider range of practical needs.

[0064] Example 2

[0065] The present invention also provides an intelligent monitoring and early warning system for aerial work, which includes: a video acquisition unit, a target detection unit, a processing unit and an early warning unit.

[0066] The video acquisition unit is used to acquire video images of the climbing preparation area, the climbing area and the high-altitude walking area in real time, and send the video images to the target detection unit.

[0067] The target detection unit is used to detect in real time whether a target, including personnel and safety equipment, appears in the video image, and send the detection result to the processing unit.

[0068] The processing unit is used to make a real-time judgment based on the detection result to determine whether there is a violation, and send the judgment result to the early warning unit.

[0069] The processing unit is used to issue an early warning in real time according to the judgment result.

[0070] Example 3

[0071] An electronic device comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein the method in the above-mentioned embodiment 1 is implemented when the processor executes the computer program.

[0072] The electronic device of the embodiment of the present application can be the mobile device itself, or a stand-alone device independent of the mobile device, which can communicate with the mobile device to receive the collected input signals from them and send the selected target decision behavior to them.

[0073] The electronic device includes one or more processors and memory. The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may run the program instructions to implement the decision behavior decision method and / or other desired functions of the various embodiments of the present application described above.

[0074] Electronic devices may also include input devices and output devices.

[0075] Example 4

[0076] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, and when the computer program instructions are executed by a processor, the processor executes the method in the above embodiment 1 of this specification.

[0077] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0078] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent monitoring and early warning method for high-altitude operations, characterized in that: Set up a climbing preparation area, a climbing area and a high-altitude walking area, and perform target detection in the three areas respectively; the targets include personnel and safety equipment; If a person is detected in the climbing preparation area, it is determined whether the person is wearing safety equipment. If the person is wearing safety equipment, the climbing preparation area is continued to be detected. If the person is not wearing safety equipment, it is determined whether the person has entered the climbing area. If the person has entered the climbing area, an early warning message is generated and archived. If the person has not entered the climbing area, the climbing preparation area is continued to be detected. If a person is detected in the climbing area, it is determined whether the person is wearing safety equipment: if the person is wearing safety equipment, the target detection in the climbing area will continue; if the person is not wearing safety equipment, an early warning message will be generated and archived; If a person is detected in the high-altitude walking area, it is determined whether the person is wearing safety equipment: if the person is wearing safety equipment, target detection of the high-altitude walking area will continue; if the person is not wearing safety equipment, a warning message will be generated and archived.

2. The intelligent monitoring and early warning method for high-altitude operations according to claim 1 is characterized in that: The safety equipment is two fall arresters; If a person is detected in the high-altitude walking area, determine whether the person is wearing a fall arrester: if the person is wearing two fall arresters, continue to perform target detection in the high-altitude walking area; if the person is not detected to be wearing two fall arresters, pause for t seconds and perform fall arrester detection again. If the person is still not detected to be wearing two fall arresters after t seconds, generate an early warning message and archive it. If the person is detected to be wearing two fall arresters after t seconds, continue to perform personnel detection and corresponding fall arrester detection in the high-altitude walking area.

3. The intelligent monitoring and early warning method for high-altitude operations according to claim 1 or 2, characterized in that: Use target detection network for target detection; At present, the detection network consists of three parts: backbone feature extraction network, PANet multi-scale feature fusion network, and feature decoder; The backbone feature extraction network is composed of a fusion downsampling module and a C2F module connected in sequence; the self-attention pyramid pooling module, namely SASPP, is introduced after the backbone feature extraction network. SASPP combines the channel attention mechanism and the spatial attention mechanism. Features of different scales are processed by SASPP and then input into the PANet multi-scale feature fusion module for enhancement; the features of different scales enhanced by the PANet multi-scale feature fusion module are respectively input into the corresponding feature decoders for decoding to identify the target.

4. The intelligent monitoring and early warning method for aerial work according to claim 3 is characterized in that: The fused downsampling module first uses the convolution layer, the maximum pooling layer and the average pooling layer in parallel to extract features from the input respectively to obtain three downsampling features, and then convolves the three downsampling features through the convolution layer respectively and then fuses them to obtain the fused downsampling features.

5. The intelligent monitoring and early warning method for aerial work according to claim 3 is characterized in that: The backbone network includes N fusion downsampling modules and N C2F modules, and the fusion downsampling modules and the corresponding C2F modules are alternately connected in sequence.

6. The intelligent monitoring and early warning method for aerial work according to claim 3 is characterized in that: The self-attention pyramid pooling module combines the channel attention mechanism and the spatial attention mechanism, specifically including a CBA module consisting of a convolutional layer, a normalization layer, and an activation function.

7. The intelligent monitoring and early warning method for aerial work according to claim 3 is characterized in that: The feature decoder adopts a decoupling strategy to process the classification task and the localization task respectively. After being processed by a CBA module, the input feature vector is divided into two branches: one branch is used to predict the confidence of the target existence and the geometric parameters of the detection box, and the other branch is used to predict the category probability of the target.

8. An intelligent monitoring and early warning system for aerial work, characterized in that: The system is applicable to the intelligent monitoring and early warning method for aerial work according to any one of claims 1 to 7, and comprises: a video acquisition unit, a target detection unit, a processing unit and an early warning unit; The video acquisition unit is used to collect video images of the climbing preparation area, the climbing area and the high-altitude walking area in real time, and send the video images to the target detection unit; The target detection unit is used to detect in real time whether a target, including personnel and safety equipment, appears in the video image, and send the detection result to the processing unit; The processing unit is used to make a real-time judgment based on the detection result to determine whether there is a violation, and send the judgment result to the early warning unit; The processing unit is used to issue an early warning in real time according to the judgment result.

9. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements an intelligent monitoring and early warning method for aerial work as described in any one of claims 1-7.

10. An electronic device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements an intelligent monitoring and early warning method for aerial work as described in any one of claims 1 to 7.

Images