Dangerous construction behavior early warning method based on image recognition

Abstract

The invention relates to a dangerous construction behavior early warning method based on image recognition, and relates to the technical field of underground cable safety protection, and the method comprises the following steps: S1, verifying a picture, S2, dividing a construction area, S3, searching the construction area, S4, determining personnel in the construction area, S5, determining personnel threats, S6, determining the type of a construction vehicle, and S7, determining whether there is personnel in the construction vehicle. S8, construction vehicle threats are judged, the advanced image recognition technology is applied, the system recognizes hidden danger vehicles and hidden danger constructors, the current common underground cable damage risk can be effectively solved, and when the system detects the potential damage risk, early warning can be given out in time, so that operation and maintenance personnel can take measures rapidly, and the construction safety is improved. Potential dangerous behaviors are interrupted, and the safety of the underground cable is ensured. Thus, through the efficient image recognition technology and the intelligent early warning system, the safety of the underground cable can be greatly improved, and the loss caused by accidental damage is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of underground cable safety protection, and in particular to an image recognition-based dangerous construction behavior early warning method. Background Art

[0002] With the increasing cabling rate in urban central areas and the accelerated implementation of new infrastructure, underground pipelines are complex, the operating environment of direct-buried high-voltage cables is becoming increasingly harsh, and the risk of damage from external forces is increasing. External damage is the main cause of cable failures. Therefore, effective measures to prevent external damage must be taken.

[0003] Since the various construction projects of urban construction and renovation come from a wide range of sources and categories, municipal construction related information and large-scale construction projects can be obtained from relevant channels, while scattered and small-scale engineering information is difficult to collect. The construction party generally does not take the initiative to connect with the cable operation and maintenance unit, so the power company cannot conduct construction briefings before construction. The construction unit blindly carries out construction work without knowing the direction and burial depth of the underground high-voltage cable, which makes it difficult to prevent cable external damage. Therefore, it is necessary to actively monitor the underground cables for external damage. The existing underground cable external damage monitoring method is to locate the external damage vibration source and identify the construction type through the same trench vibration monitoring method. Although it can achieve certain results, it cannot realize the visualization of the construction site and the verification of the construction position, and the reliability is not high. Faced with increasingly complex construction site conditions, in order to improve the limitations of the underground cable external damage identification algorithm, it is necessary to optimize and upgrade the external damage intelligent algorithm. To this end, we propose a dangerous construction behavior early warning method based on image recognition. Summary of the invention

[0004] The purpose of this invention is to overcome the above technical problems and provide a dangerous construction behavior warning method based on image recognition. It uses advanced image recognition technology to effectively identify the current common underground cable damage risks so that operation and maintenance personnel can quickly take measures to interrupt potential dangerous behaviors and ensure the safety of underground cables. Through efficient image recognition technology and intelligent warning systems, the safety of underground cables can be greatly improved and the losses caused by accidental damage can be reduced.

[0005] The present invention solves the technical problem by the following technical solutions:

[0006] A dangerous construction behavior early warning method based on image recognition includes the following steps:

[0007] S1. Verify the image, use image acquisition equipment to collect ground images of underground cable routes, and then use image recognition algorithms to identify roads, barriers, and river barriers;

[0008] S2. Divide the construction area. After using satellite positioning equipment to locate the cable position, mark the cable position and construction area on the collected image to divide the underground cable and the construction area;

[0009] S3, search the construction area, mark the location of the underground cables at the construction site, and then use the image recognition algorithm to determine whether there are earth bags, construction vehicles and construction personnel above the cable location, and make subsequent hidden danger determinations when there is a construction area above the underground cables;

[0010] S4. Determine the personnel in the construction area. If there are construction personnel, use the image recognition algorithm to identify whether the personnel are in the construction area;

[0011] S5. Determine the threat of personnel and identify whether the construction personnel are carrying digging equipment. If the construction personnel are not carrying digging equipment, the hidden danger determination is terminated. After the digging equipment is identified, a comprehensive determination is made on the construction personnel and the digging equipment. The images collected in two cycles are compared to determine whether the construction personnel have a digging tendency. If there is a digging tendency, a hidden danger warning is output;

[0012] S6. Determine the type of construction vehicles, use image recognition algorithms to identify the detection model above the underground cables, and distinguish the types of vehicles for the detection model, including the identification of cars, trucks, excavators and pile drivers, and determine that excavators and pile drivers are types of vehicles with hidden dangers;

[0013] S7, determining whether there are people in the construction vehicle, and after determining that it is a potential hazard vehicle, using an image recognition algorithm to obtain a vehicle model, and then cutting the vehicle model to identify whether there are people in the cockpit in the vehicle driving area;

[0014] S8. Determine the threat of construction vehicles. If no one is found in the potential danger vehicle, the potential danger determination is terminated. If there is a person in the potential danger vehicle and it is impossible to determine due to reflection or lack of light, in the next image acquisition cycle, the image recognition algorithm is used to compare the potential danger vehicle with the image of the previous cycle to determine whether the potential danger vehicle is moving. If there is a tendency to dig the ground, a potential danger warning is output.

[0015] Furthermore, the image recognition algorithm used in S1-S8 is a convolutional neural network. The pre-processed data of the images collected by the image acquisition device is substituted into the convolutional neural network for convolution operations, activation functions, pooling operations and calculations of the fully connected layer. The trained convolutional neural network is then deployed into the production environment and input into the convolutional neural network model for prediction.

[0016] Furthermore, the calculation formula for extracting local features by convolution operation is:

[0017]

[0018] Among them, I(i, j) is the pixel value of the input feature map at position (i, j), K(m, n) is the weight of the convolution kernel at position (m, n), O(i, j) is the pixel value of the output feature map at position (i, j), and M and N are the height and width of the convolution kernel, respectively.

[0019] Furthermore, the calculation formula of the activation function is:

[0020] f(x) = max(0, x)

[0021] This is the activation function of ReLU, where f(x) represents the output of the ReLU activation function and x represents the input value.

[0022] Furthermore, the pooling operation adopts the maximum pooling, and the maximum pooling calculation formula is,

[0023]

[0024] Among them, O(i, j) is the pixel value of the output feature map at position (i, j), I is the input feature map, p×p is the pooling window size, and I(ip+m,jp+n) represents the local area in the input feature map.

[0025] Furthermore, the calculation formula of the fully connected layer is,

[0026] y=Wx+b

[0027] Where x is the input vector, W is the weight matrix, and b is the bias vector.

[0028] Furthermore, the loss of the predicted value calculated by the fully connected layer can be calculated using the cross entropy loss. The calculation formula of the cross entropy loss is:

[0029]

[0030] Where N is the number of samples, y i is the true label of the i-th sample, is the predicted probability of the ith sample.

[0031] Furthermore, the prediction parameters need to be updated by the chain rule, and the calculation formula is:

[0032]

[0033] Among them, L is the loss function and θ is the model parameter.

[0034] The advantages and positive effects of the present invention are:

[0035] The present invention uses advanced image recognition technology to accurately classify and identify scenes such as roads, fences, and rivers. After completing the classification, the system will further divide the positions between underground cables and fences in order to better identify potential dangerous factors. The system can effectively solve the current common underground cable damage risks by identifying potential vehicles and potential construction personnel. The system can also detect whether there is an obstruction between the two, and whether there is construction behavior between construction personnel and potential vehicles. When the system detects potential damage risks, it will issue an early warning in time so that operation and maintenance personnel can quickly take measures to interrupt potential dangerous behaviors and ensure the safety of underground cables. In this way, through efficient image recognition technology and intelligent early warning systems, the safety of underground cables can be greatly improved and the losses caused by accidental damage can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the dangerous construction behavior early warning method based on image recognition of the present invention; DETAILED DESCRIPTION

[0037] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0038] See also Figure 1 The present invention provides a dangerous construction behavior early warning method based on image recognition, comprising the following steps:

[0039] S1. Verify the image, use image acquisition equipment to collect ground images of underground cable routes, and then use image recognition algorithms to identify roads, barriers, and river barriers;

[0040] S2. Divide the construction area. After using satellite positioning equipment to locate the cable position, mark the cable position and construction area on the collected image to divide the underground cable and the construction area;

[0041] S3, search the construction area, mark the location of the underground cables at the construction site, and then use the image recognition algorithm to determine whether there are earth bags, construction vehicles and construction personnel above the cable location, and make subsequent hidden danger determinations when there is a construction area above the underground cables;

[0042] Determining whether a vehicle or worker is a construction vehicle usually involves the following criteria:

[0043] Image feature recognition: Construction workers are required to wear uniforms and carry tools. They usually wear uniform work clothes, such as reflective vests, helmets, protective glasses, safety shoes, etc., and carry construction tools, such as shovels, pickaxes, and measuring tools. Construction vehicle identification includes vehicle type and vehicle paint, and identifies different types of construction vehicles, such as excavators, bulldozers, and concrete mixers. Further segmentation is performed based on the characteristics of vehicle paint.

[0044] For example, some construction vehicles are painted with unique logos and patterns, which often represent the company or project to which the vehicle belongs. Through image recognition technology, the system can quickly match the paint features in the database to achieve accurate identification and management of vehicles. This not only improves the management efficiency of the construction site, but also enhances safety monitoring capabilities.

[0045] Behavioral feature recognition: Construction workers will perform construction activities such as digging, carrying materials, measuring, and directing near the cable location. Through image recognition algorithms, the postures of personnel, such as bending over to dig, swinging tools, etc., the movement paths of construction vehicles, and the construction actions of working equipment can be analyzed.

[0046] Position relationship: the distance between construction vehicles and construction workers and the cable location.

[0047] S4. Determine the personnel in the construction area. If there are construction personnel, use the image recognition algorithm to identify whether the personnel are in the construction area;

[0048] S5. Determine the threat of personnel and identify whether the construction personnel are carrying digging equipment. If the construction personnel are not carrying digging equipment, the hidden danger determination is terminated. After the digging equipment is identified, a comprehensive determination is made on the construction personnel and the digging equipment. The images collected in two cycles are compared to determine whether the construction personnel have a digging tendency. If there is a digging tendency, a hidden danger warning is output;

[0049] S6. Determine the type of construction vehicles, use image recognition algorithms to identify the detection model above the underground cables, and distinguish the types of vehicles for the detection model, including the identification of cars, trucks, excavators and pile drivers, and determine that excavators and pile drivers are types of vehicles with hidden dangers;

[0050] S7, determining whether there are people in the construction vehicle, and after determining that it is a potential hazard vehicle, using an image recognition algorithm to obtain a vehicle model, and then cutting the vehicle model to identify whether there are people in the cockpit in the vehicle driving area;

[0051] S8. Determine the threat of construction vehicles. If no one is found in the potential danger vehicle, the potential danger determination is terminated. If there is a person in the potential danger vehicle and it is impossible to determine due to reflection or lack of light, in the next image acquisition cycle, the image recognition algorithm is used to compare the potential danger vehicle with the image of the previous cycle to determine whether the potential danger vehicle is moving. If there is a tendency to dig the ground, a potential danger warning is output.

[0052] In this embodiment, preferably, the image recognition algorithm used in S1-S8 is a convolutional neural network, and the pre-processed data of the pictures collected by the image acquisition device is substituted into the convolutional neural network to perform convolution operations, activation functions, pooling operations and calculations of the fully connected layer, and then the trained convolutional neural network is deployed into the production environment and input into the convolutional neural network model for prediction.

[0053] In this embodiment, preferably, the calculation formula for extracting local features by convolution operation is:

[0054]

[0055] Among them, I(i, j) is the pixel value of the input feature map at position (i, j), K(m, n) is the weight of the convolution kernel at position (m, n), O(i, j) is the pixel value of the output feature map at position (i, j), and M and N are the height and width of the convolution kernel, respectively.

[0056] In this embodiment, preferably, the calculation formula of the activation function is:

[0057] f(x) = max(0, x)

[0058] This is the activation function of ReLU, where f(x) represents the output of the ReLU activation function and x represents the input value.

[0059] In this embodiment, preferably, the pooling operation adopts maximum pooling, and the maximum pooling calculation formula is:

[0060]

[0061] Among them, O(i, j) is the pixel value of the output feature map at position (i, j), I is the input feature map, p×p is the pooling window size, and I(ip+m,jp+n) represents the local area in the input feature map.

[0062] In this embodiment, preferably, the calculation formula of the fully connected layer is:

[0063] y=Wx+b

[0064] Where x is the input vector, W is the weight matrix, and b is the bias vector.

[0065] In this embodiment, preferably, the loss of the predicted value calculated by the fully connected layer can be calculated using the cross entropy loss, and the calculation formula of the cross entropy loss is:

[0066]

[0067] Where N is the number of samples, y i is the true label of the i-th sample, is the predicted probability of the ith sample.

[0068] In this embodiment, preferably, the prediction parameter needs to be updated by the chain rule, and its calculation formula is:

[0069]

[0070] Among them, L is the loss function, y is the prediction parameter, and θ is the model parameter.

[0071] The working principle and use process of the present invention: When the device is used, the image acquisition device uses an online monitoring system with 24-hour centralized monitoring, and performs data sorting and operation and maintenance guarantee through the intelligent control center. By using advanced image recognition technology, it can accurately classify and identify scenes such as roads, enclosures, rivers, etc. After completing the classification, the system will further divide the position between the underground cable and the enclosure to better identify potential dangerous factors. The system can effectively solve the current common underground cable damage risks by identifying hidden danger vehicles and hidden danger construction personnel. For example, the system can detect whether there are construction personnel or hidden danger vehicles within a certain distance above the underground cable path, and can also determine whether the construction personnel and hidden danger vehicles have entered the potential danger area of ​​the underground cable. In addition, the system can also detect whether there is an obstruction between the two, and whether there is construction behavior between the construction personnel and the hidden danger vehicle. When the system detects a potential risk of damage, it will issue an early warning in time so that the operation and maintenance personnel can quickly take measures to interrupt the potential dangerous behavior and ensure the safety of the underground cable. In this way, through efficient image recognition technology and intelligent early warning system, the safety of underground cables can be greatly improved and the losses caused by accidental damage can be reduced.

[0072] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A dangerous construction behavior early warning method based on image recognition, characterized in that: It includes the following steps: S1. Verify the image, use image acquisition equipment to collect ground images of underground cable routes, and then use image recognition algorithms to identify roads, barriers, and river barriers; S2. Divide the construction area. After using satellite positioning equipment to locate the cable position, mark the cable position and construction area on the collected image to divide the underground cable and the construction area; S3, search the construction area, mark the location of the underground cables at the construction site, and then use the image recognition algorithm to determine whether there are earth bags, construction vehicles and construction personnel above the cable location, and make subsequent hidden danger determinations when there is a construction area above the underground cables; S4. Determine the personnel in the construction area. If there are construction personnel, use the image recognition algorithm to identify whether the personnel are in the construction area; S5. Determine the threat of personnel and identify whether the construction personnel are carrying digging equipment. If the construction personnel are not carrying digging equipment, the hidden danger determination is terminated. After the digging equipment is identified, a comprehensive determination is made on the construction personnel and the digging equipment. The images collected in two cycles are compared to determine whether the construction personnel have a digging tendency. If there is a digging tendency, a hidden danger warning is output; S6. Determine the type of construction vehicles, use image recognition algorithms to identify the detection model above the underground cables, and distinguish the types of vehicles for the detection model, including the identification of cars, trucks, excavators and pile drivers, and determine that excavators and pile drivers are types of vehicles with hidden dangers; S7, determining whether there are people in the construction vehicle, and after determining that it is a potential hazard vehicle, using an image recognition algorithm to obtain a vehicle model, and then cutting the vehicle model to identify whether there are people in the cockpit in the vehicle driving area; S8. Determine the threat of construction vehicles. If no one is found in the potential danger vehicle, the potential danger determination is terminated. If there is a person in the potential danger vehicle and it is impossible to determine due to reflection or lack of light, in the next image acquisition cycle, the image recognition algorithm is used to compare the potential danger vehicle with the image of the previous cycle to determine whether the potential danger vehicle is moving. If there is a tendency to dig the ground, a potential danger warning is output.

2. According to claim 1, a dangerous construction behavior early warning method based on image recognition is characterized by: The image recognition algorithm used in S1-S8 is a convolutional neural network. The pre-processed data of the images collected by the image acquisition device is substituted into the convolutional neural network for convolution operation, activation function, pooling operation and calculation of the fully connected layer. The trained convolutional neural network is then deployed into the production environment and input into the convolutional neural network model for prediction.

3. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The calculation formula for extracting local features by convolution operation is: Among them, I(i, j) is the pixel value of the input feature map at position (i, j), K(m, n) is the weight of the convolution kernel at position (m, n), O(i, j) is the pixel value of the output feature map at position (i, j), and M and N are the height and width of the convolution kernel, respectively.

4. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The calculation formula of the activation function is: f(x) = max(0, x) This is the activation function of ReLU, where f(x) represents the output of the ReLU activation function and x represents the input value.

5. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The pooling operation uses maximum pooling, and the maximum pooling calculation formula is: Among them, O(i, j) is the pixel value of the output feature map at position (i, j), I is the input feature map, p×p is the pooling window size, and I(ip+m,jp+n) represents the local area in the input feature map.

6. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The calculation formula of the fully connected layer is: v=Wx+b Where x is the input vector, W is the weight matrix, and b is the bias vector.

7. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The loss of the predicted value calculated by the fully connected layer can be calculated using the cross entropy loss. The calculation formula of the cross entropy loss is: Where N is the number of samples, y i is the true label of the i-th sample, is the predicted probability of the ith sample.

8. The method for early warning of dangerous construction behaviors based on image recognition according to claim 2 is characterized by: The prediction parameters need to be updated by the chain rule, and the calculation formula is: Among them, L is the loss function and θ is the model parameter.

Images