Inspection Method, System and Storage Medium for Construction Environment

Abstract

The present invention discloses a method, a system and a storage medium for inspecting a construction environment. In each inspection cycle, the construction risk coefficient, the personnel location information and the camera information of at least one target person in the construction environment are obtained, and then these data are input into a pre-trained inspection planning model to output the inspection probability of each target person. The target person corresponding to the maximum value among all the inspection probabilities is selected as the inspection target for the current inspection cycle. Then, in the current inspection cycle, the focusing direction of the target camera is controlled to be aligned with the inspection target until all inspection cycles are completed. By adopting the embodiment of the present invention, the inspection resources can be utilized most effectively, blind inspection is avoided, and the pertinence and efficiency of inspection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of big data processing, and particularly relates to a patrol inspection method, system and storage medium for a construction environment. Background Art

[0002] The existing safety supervision commonly uses the manual + camera patrol inspection mode. The common safety patrol inspection process is as follows: the construction contracting team enters the corresponding information at the construction site according to the work permit, and installs cameras that can be connected to the safety supervision center. The duty personnel in the monitoring center select the construction site, remotely control the cameras to rotate and zoom in on the monitoring screen to conduct construction safety compliance inspections. This patrol inspection method will consume a large amount of labor costs. Since the individual risk differences of construction personnel are not fully considered, the patrol inspection resources cannot be effectively utilized. It may result in insufficient patrol inspection of some high-risk personnel, while unnecessary and blind patrol inspection of some low-risk personnel, leading to some potential safety hazards not being discovered and processed in a timely manner. Summary of the Invention

[0003] The purpose of the embodiments of the present invention is to provide a patrol inspection method, system and storage medium for a construction environment, which can make the patrol inspection resources be utilized most effectively, avoid blind patrol inspection, and improve the pertinence and efficiency of patrol inspection.

[0004] To achieve the above purpose, the embodiments of the present invention provide a patrol inspection method for a construction environment, including:

[0005] In each patrol inspection cycle, obtain the construction risk coefficients and personnel location information of all target personnel in the construction environment, and obtain the camera information of the target cameras in the construction environment;

[0006] Input the construction risk coefficients, the personnel location information and the camera information into the patrol inspection planning model, so that the patrol inspection planning model outputs the patrol inspection probability of each target personnel;

[0007] Select the target personnel corresponding to the maximum value from all the patrol inspection probabilities as the patrol inspection target for this patrol inspection cycle;

[0008] In this patrol inspection cycle, control the focusing direction of the target camera to be aligned with the patrol inspection target until all patrol inspection cycles are completed.

[0009] As an improvement of the above solution, the construction risk coefficient is calculated according to the risk degree value corresponding to the personnel information in the target personnel; wherein, the personnel information includes at least one of basic information, team information and construction information, and each personnel information corresponds to a risk degree value.

[0010] As an improvement to the above solution, the basic information includes attribute information and first historical violation information; when the personnel information includes basic information, the method for determining the first risk degree value corresponding to the basic information includes:

[0011] Divide the historical data into positive and negative samples; among them, the positive samples include data records without safety accidents, and the negative samples include data records with safety accidents;

[0012] For the attribute information in the negative samples, use the clustering algorithm for clustering analysis to obtain several attribute categories;

[0013] Calculate the contribution rate of each attribute category to the occurrence of safety accidents, and integrate the contribution rates of all attribute categories to obtain a contribution rate set;

[0014] Obtain the target contribution rate matching the target personnel from the contribution rate set, and obtain the violation risk coefficient corresponding to the first historical violation information;

[0015] Use the sum of the target contribution rate and the violation risk coefficient as the first risk degree value.

[0016] As an improvement to the above solution, calculating the contribution rate of each attribute category to the occurrence of safety accidents includes:

[0017] For each attribute category, count the number of positive samples belonging to the current attribute category in the positive samples, and count the number of negative samples belonging to the current attribute category in the negative samples;

[0018] Determine the contribution rate of the current attribute category to the occurrence of safety accidents according to the number of positive samples and the number of negative samples.

[0019] As an improvement to the above solution, the first historical violation information includes the first time interval from the violation time of the target personnel to the current time, and the violation risk coefficient is inversely proportional to the first time interval.

[0020] As an improvement to the above solution, the team information includes second historical violation information and historical working hours, and the second historical violation information is the second time interval from the violation time of any person in the team to the current time; when the personnel information includes team information, the method for determining the second risk degree value corresponding to the team information includes:

[0021] Calculate the first risk coefficient according to the second time interval, and calculate the second risk coefficient according to the historical working hours; among them, the second time interval and the first risk coefficient are inversely proportional, and the historical working hours and the second risk coefficient are inversely proportional;

[0022] The sum of the first risk coefficient and the second risk coefficient is taken as the second risk degree value.

[0023] As an improvement of the above solution, the construction information includes the construction type and its corresponding construction risk level; when the personnel information includes construction information, the third risk level value corresponding to the construction information is directly proportional to the construction risk level.

[0024] As an improvement of the above scheme, the inspection planning model is a reinforcement learning model, a reward function is provided in the inspection planning model, and the inspection planning model takes the maximum value of the reward function output as the optimization target; wherein the reward function is related to the construction risk coefficient.

[0025] To achieve the above purpose, an embodiment of the present invention further provides a construction environment inspection system, comprising:

[0026] An information acquisition module is used to obtain the construction risk factor and personnel location information of all target personnel in the construction environment in each inspection cycle, and to obtain the camera information of the target camera in the construction environment;

[0027] An inspection probability generation module, used for inputting the construction risk coefficient, the personnel location information and the camera information into the inspection planning model, so that the inspection planning model outputs the inspection probability of each target personnel;

[0028] Inspection target determination module, used to select the target personnel corresponding to the maximum value from all inspection probabilities as the inspection target of this inspection cycle;

[0029] The inspection control module is used to control the focus direction of the target camera to align with the inspection target in this inspection cycle until all inspection cycles are completed.

[0030] To achieve the above-mentioned purpose, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the inspection method of the construction environment as described in any of the above-mentioned embodiments.

[0031] Compared with the prior art, the inspection method, system, and storage medium for the construction environment disclosed in the present invention can accurately select the person with the highest risk as the inspection target from among many uninspected persons by obtaining the construction risk coefficient of the target person, combining the personnel location information and camera information, and using the inspection planning model to output the inspection probability. This enables the most effective utilization of inspection resources, avoids blind inspections, and improves the pertinence and efficiency of inspections. When determining the inspection target, the personnel location information and camera information are also considered, which helps to plan a more reasonable inspection path, reduce time waste and distance consumption during the inspection process, improve the overall inspection efficiency, and shorten the time required to complete all inspection cycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is the system topology diagram of the inspection method applied in the embodiment of the present invention;

[0033] Figure 2 is the flowchart of an inspection method for a construction environment provided by an embodiment of the present invention;

[0034] Figure 3 is another flowchart of an inspection method for a construction environment provided by an embodiment of the present invention;

[0035] Figure 4 is the schematic diagram of the positioning device for positioning the target person provided by an embodiment of the present invention;

[0036] Figure 5 is the working flowchart of the inspection planning model provided by an embodiment of the present invention;

[0037] Figure 6 is the structural block diagram of an inspection system for a construction environment provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0039] See Figure 1 , Figure 1 is the system topology diagram of the inspection method applied in the embodiment of the present invention. It can be divided into three layers from bottom to top, corresponding to the intelligent inspection device for the power construction site (field layer), the Internet (transmission layer), and the safety management platform layer. In the field layer, it mainly includes an edge computing device, a camera, an audible and visual alarm, and a positioning device.

[0040] The edge computing device is used to provide computing and storage capabilities at the infrastructure construction site, deploy the trained reinforcement learning inference model, and process input data including but not limited to video data uploaded by cameras, crew member data, sensing data, etc. Based on this data, it determines inspection targets, executes inspection strategies, etc., and uploads inspection images that meet the violation recognition accuracy to the safety management platform, while receiving alarm instructions issued by the management platform.

[0041] The camera is the control object of the reinforcement learning model. First, it can upload its own device status (including the focus angle and focal length of the current camera) and the video frame data it captures. Second, it can receive and execute control commands (including rotating the camera and zooming).

[0042] The audible and visual alarm is connected to the edge computing device by wire or wirelessly. When the violation recognition model of the safety management platform detects a violation construction behavior at the construction site, it will issue an alarm instruction to the edge computing device, which will then trigger the audible and visual alarm function of the audible and visual alarm.

[0043] The positioning device can accurately locate the personnel in the construction environment. For example, if the positioning device is a UWB (Ultra WideBand) positioning system, it includes two parts: a UWB base station and a UWB tag. The UWB tag needs to be worn in the hands of the target personnel, and at the same time, 4 UWB base stations need to be deployed at the construction site to achieve ranging and positioning based on TDOA (Time Difference of Arrival).

[0044] It should be noted that during the system initialization phase (such as when construction starts every day), the above devices need to be checked for health and configuration loaded. Among them, the health check mainly includes the network link status of the edge computing device connected to the safety management platform, the computing / storage status of the edge computing device, the camera status, the positioning device, and the network link status between the camera, the positioning device and the edge computing device, whether the audible and visual alarm can work properly, etc. The configuration loading is mainly to load some algorithm models, such as the inspection planning model, etc.

[0045] See Figure 2 , Figure 2 is a flowchart of an inspection method for a construction environment provided by an embodiment of the present invention. The inspection method for the construction environment is implemented by an edge computing device, and the inspection method for the construction environment includes:

[0046] S1. In each inspection cycle, obtain the construction risk coefficients and personnel location information of all target personnel in the construction environment, and obtain the camera information of the target cameras in the construction environment;

[0047] S2. Input the construction risk coefficient, the personnel location information, and the camera information into the patrol inspection planning model, so that the patrol inspection planning model outputs the patrol inspection probability of each target personnel.

[0048] S3. Select the target personnel corresponding to the maximum value from all the patrol inspection probabilities as the patrol inspection target for this patrol inspection cycle.

[0049] S4. During this patrol inspection cycle, control the focusing direction of the target camera to be aligned with the patrol inspection target until all patrol inspection cycles have completed the patrol inspection.

[0050] Exemplarily, deploy the patrol inspection planning model on an edge computing device. The patrol inspection planning model can determine the target personnel that need to be key monitored during this patrol inspection cycle, so as to control the target camera deployed on site to periodically patrol the construction status of the target personnel. It can not only automatically complete the lens rotation, but also make the occupancy of the target personnel in the captured image exceed 50% through automatic docking, and transmit it back to the security management platform deployed in the cloud to complete the detection of illegal behaviors. In addition, due to different work categories and uneven comprehensive qualities of the team members at the construction site, the safety risks of each target personnel are different. The present invention will calculate the risk coefficient of each target personnel according to the personnel information of the target personnel, and based on the location of each target personnel, use the patrol inspection planning model to calculate the overall risk cost to be paid for personnel patrol inspection within one cycle. The ultimate goal is to obtain the maximum global benefit (i.e., the overall risk cost is the smallest), and then screen out the patrol inspection targets with higher risk levels for targeted patrol inspection.

[0051] See Figure 3 , Figure 3 is another flow chart of a patrol inspection method for a construction environment provided by an embodiment of the present invention. In combination with Figure 3 the above steps S1 to S4 are described in detail.

[0052] Specifically, in step S1, the construction risk coefficient is calculated according to the risk degree value corresponding to the personnel information among the target personnel; wherein, the personnel information includes at least one of basic information, team information, and construction information, and each personnel information corresponds to a risk degree value. The detailed content of each personnel information is as follows:

[0053] 1) Basic information, the basic information includes attribute information and the first historical violation information. The basic information part represents the basic physical quality of the construction personnel, and to a certain extent reflects the strength of the safety construction awareness of the construction personnel through historical violations. The attribute information includes gender, age, working years, etc., and each piece of data in the first historical violation information includes three dimensions, namely the violation time (the first time interval between the violation time of the target personnel and the current time), the severity, and the violation content.

[0054] 2) Team information, where the team information includes second historical violation information and historical working hours. The team information reflects the management level of the team to which the target person belongs. The second historical violation information also includes three dimensions: violation time (the second time interval from the violation time of any person in the team to the current time), severity, and violation content. The historical working hours represent the degree of experience of the team.

[0055] 3) Construction information, where the construction information includes the construction type and its corresponding construction risk level, and details the environmental status faced by the target person in this operation. The construction types include welding, cutting, wire laying, etc., and each has a corresponding construction risk level. In addition, the construction risk level can further consider the influence of factors such as construction time, construction location, and meteorological indicators.

[0056] For the above three types of personnel information, the determination process of their corresponding risk degree values is described separately:

[0057] Case 1: When the personnel information includes basic information, the method for determining the first risk degree value corresponding to the basic information includes: dividing the historical data into positive and negative samples; where the positive samples include data records without safety accidents, and the negative samples include data records with safety accidents; for the attribute information in the negative samples, use the clustering algorithm for clustering analysis to obtain several attribute categories; calculate the contribution rate of each attribute category to the occurrence of safety accidents, and integrate the contribution rates of all attribute categories to obtain a contribution rate set; obtain the target contribution rate matching the target person from the contribution rate set, and obtain the violation risk coefficient corresponding to the first historical violation information; use the sum of the target contribution rate and the violation risk coefficient as the first risk degree value. Among them, calculating the contribution rate of each attribute category to the occurrence of safety accidents includes: for each attribute category, count the number of positive samples belonging to the current attribute category in the positive samples, and count the number of negative samples belonging to the current attribute category in the negative samples, and determine the contribution rate of the current attribute category to the occurrence of safety accidents according to the number of positive samples and the number of negative samples.

[0058] Exemplarily, all historical data is divided into positive and negative samples according to whether a safety accident has occurred, that is , where represents the positive sample, that is, the data record without a safety accident, represents the negative sample, that is, the data record with a safety accident. For The attribute information (gender, age and length of service) in the basic information in the set is clustered and analyzed using a clustering algorithm. For example, the clustering algorithm used in the embodiment of the present invention is the K-Means algorithm, and the k value is determined by the silhouette coefficient method. The k value represents the number of clusters into which the data is to be divided. The attribute information of is represented by a data to be clustered, such as gender, age, and length of service, which can be expressed as , assuming that the total amount of data of the target person is n, satisfying , then the value range of k is set to .

[0059] Iterate clustering for different k values ​​and use the following formula to calculate the target personnel The silhouette coefficient of the data to be clustered:

[0060] (1);

[0061] in, The target person The average distance between the data to be clustered (sample points) and all other points in its cluster, The target person The average distance between the data to be clustered (sample points) and all the points in the nearest cluster. For a k value, the silhouette coefficient satisfy:

[0062] (2);

[0063] but , that is, take the k value that minimizes the silhouette coefficient. So far, the data to be clustered is divided into k categories through the k-means algorithm, which is recorded as .

[0064] Then, the contribution rate of each cluster to whether an accident can be judged is traversed, recorded as , the calculation process of the target personnel's contribution rate is as follows:

[0065] 1. Statistics The attribute category in the collection The number of records is recorded as ;

[0066] 2. Statistics The attribute category in the collection The number of records is recorded as ;

[0067] 3. Calculate contribution rate ,satisfy:

[0068] (3);

[0069] 4. Traverse , repeat the above process to obtain .

[0070] Finally, integrate the contribution rates of all attribute categories to obtain a contribution rate set , and obtain the target contribution rate matching the target person from the contribution rate set. During the screening process, for each target person, traverse , find the most similar attribute category. For example, obtain a similarity index through a similarity calculation method, select the target attribute category corresponding to the maximum value, and then index the contribution rate corresponding to the target attribute category as the target contribution rate.

[0071] Exemplarily, after obtaining the target contribution rate of the target person and obtaining the violation risk coefficient corresponding to the first historical violation information, use the sum of the target contribution rate and the violation risk coefficient as the first risk level value, satisfying the following formula:

[0072] (4);

[0073] Wherein, is the target contribution rate of the target person , that is, the risk coefficient representing the clustering category to which it belongs; is the target person 's violation risk coefficient.

[0074] Specifically, the first historical violation information includes the first time interval between the violation time of the target person and the current time, and the violation risk coefficient is inversely proportional to the first time interval.

[0075] Exemplarily, use the idea of attenuation to calculate the proportion of historical violations. Assume that the first time interval between the historical violation time of the target person and the current month is expressed as , represents the month after time passes, then the violation risk coefficient of the target person satisfies:

[0076] (5).

[0077] Case 2: When the personnel information includes team information, the method for determining the second risk degree value corresponding to the team information includes: calculating a first risk coefficient according to the second time interval, and calculating a second risk coefficient according to the historical working hours; wherein, the second time interval and the first risk coefficient are inversely proportional, and the historical working hours and the second risk coefficient are inversely proportional; the sum of the first risk coefficient and the second risk coefficient is used as the second risk degree value.

[0078] Exemplarily, for the target personnel the same calculation method as formula (5) in Case 1 is adopted for the second historical violation information. The reciprocal sum of the month spans is used as the first risk coefficient of this part, denoted as . For the team historical work, the reciprocal of the cumulative months is used as the second risk coefficient of this part, denoted as . Then, for the target personnel the second risk degree value corresponding to the team information satisfies:

[0079] (6);

[0080] Furthermore, in order to be consistent with the order of magnitude of the basic information part, the second risk degree values of all target personnel in this construction can be normalized. The calculation process of data normalization can refer to the prior art, and the present invention does not make specific limitations on this.

[0081] Case 3: When the personnel information includes construction information, the third risk degree value corresponding to the construction information is directly proportional to the construction risk level.

[0082] Exemplarily, in the same way as the sample division method in Case 1, count the number of all construction types in the negative sample set. For all construction types that have experienced safety accidents, sort them in descending order according to their quantities to obtain the construction risk values corresponding to different construction types . These construction risk values can be further normalized. The construction risk levels are divided into four categories: major risk, relatively large risk, general risk, and low risk. The construction risk levels correspond to different construction types and are used as the weights of the construction types, which can be taken according to empirical values. For example, the construction risk level is denoted as . After obtaining the construction type of the target personnel, multiply the construction risk value and the construction risk level to obtain the third risk degree value corresponding to the construction information. Then, for the target personnel the third risk degree value satisfies:

[0083] (7).

[0084] Further, in order to be consistent with the order of magnitude of the basic information part, a normalization calculation is performed on the operation risks of all personnel in this construction. The calculation process of data normalization can refer to the prior art, and the present invention does not make specific limitations thereon.

[0085] Further, assume that at this time the target personnel has the above three types of information (basic information, team information, and construction information). Then, the construction risk coefficient of the target personnel is obtained by adding the risk degree values corresponding to the three types of information, such as satisfying the following formula:

[0086] (8).

[0087] Specifically, a positioning device is provided on each target personnel, and the personnel position information of each target personnel is determined by receiving the positioning information of the positioning device.

[0088] Exemplarily, referring to Figure 4 , Figure 4 is a schematic diagram of the positioning device provided by the embodiment of the present invention for positioning the target personnel. The acquisition of the target personnel position is realized through a UWB positioning system, which includes two parts: a UWB base station and a UWB tag. Four UWB base stations are deployed at the construction site, and positioning is performed based on TDOA using the time difference of arrival. The UWB tag worn by the target personnel sends out a UWB signal once, and all UWB base stations within the wireless coverage range of the tag will receive the wireless signal. If two UWB base stations with known coordinate points receive the signal, and the intervals between the UWB tag and the two UWB base stations are different, then the time points when the two UWB base stations receive the signal are different. Therefore, the concept of a "time difference of arrival" is obtained.

[0089] According to the above principle, assume that the three-dimensional coordinates of the target personnel are , and the positions of the four UWB base stations are respectively , , , . It is known that the signal propagation speed is a determined value denoted as , and the times when the tag signal arrives at the four base stations are respectively denoted as , , , , and the distances from the four base stations are , , , . Use to represent the target personnel The distance difference from base stations m and n yields the following formula:

[0090] (9);

[0091] wherein, 、 、 、 and the propagation speed are known scalars, and the result of Equation (9) can be directly calculated.

[0092] Furthermore, since the coordinates of 4 base stations are known, the following system of equations can be obtained, and by solving it, the three-dimensional coordinates of the target person can be obtained. .

[0093] (10).

[0094] Thus, the solution of the three-dimensional coordinates of the target person based on the UWB positioning system is completed.

[0095] Specifically, in step S2, the inspection planning model is a reinforcement learning model, and a reward function is provided in the inspection planning model. The inspection planning model aims to optimize the maximum value output by the reward function; wherein, the reward function is related to the construction risk coefficient.

[0096] Exemplarily, referring to Figure 5 , Figure 5 is the workflow diagram of the inspection planning model provided by the embodiment of the present invention. The inspection planning model is a DDQN (Double Deep Q-Network) model, represents the environmental state at time t and is the input of the reinforcement learning model. includes the three-dimensional coordinates of each target person (such as A and B in the figure) and the construction risk coefficient . Further, it can also include a mark indicating whether the target person has completed the inspection. The existence of this mark can avoid repeatedly selecting the same target person as the inspection target. If the target person A has completed the inspection, it is marked as 1, otherwise 0. This inspection mark can be denoted as . In addition, also includes the camera information of the target camera, such as the three-dimensional coordinates of the target camera and the direction vector of the current focusing direction . Therefore, the state description of the infrastructure construction site can be denoted as:

[0097] .

[0098] It should be noted that at the initial stage of the task, the spatial position coordinates of the target camera are camera_position=(0,0,0), and the unit spatial vector corresponding to the initial field of view center of the target camera is camera_focus=(0,0,1). In the embodiments of the present invention, a target camera can be arranged specifically for monitoring the inspection target. Of course, there can also be multiple cameras in a construction environment. The other cameras except the target camera work normally according to the set program.

[0099] During the execution of the task, the physical position coordinates of the target camera remain unchanged. The actions that the target camera can perform include camera rotation and autofocus. Rotation aligns the center of the lens field of view with the target to be collected, and adjusts the lens focal length to a suitable position so that the target to be photographed occupies more than 50% of the frame in the collected image. The action space of the target camera is the target personnel who have not been inspected during this inspection cycle.

[0100] It should be noted that when the inspection task is started, the target camera is located at the initialization position. When the th inspection target is inspected, the direction vector of the center of the target camera's field of view is updated to . At this time, the inspection of the next target is carried out. The actions required at this time consist of two parts: one is the rotation of the target camera: record the required time , which is the ratio of the included angle between and to the angular velocity of the target camera's rotation. The other is to change the focal length: record the required time . The length of the focusing time is proportional to the distance between the target camera and the inspection target, and the value is the ratio of the distance between the inspection target and the coordinate origin to the focusing speed of the target camera.

[0101] Furthermore, in the problem scenario to be solved by the present invention, since the waiting time in the previous stage, all inspection targets may have risks starting from the inspection task. The probability of occurrence of risks is proportional to their risk coefficients and waiting times. Use to represent the inspection duration required for inspecting the th inspection target, to represent the construction risk coefficient of the th inspection target, to represent the risk cost paid when the th target is inspected. For waiting for an inspection target to complete the inspection, the risk cost to be paid satisfies:

[0102] (11);

[0103] Therefore, the overall risk cost from the start to the end of the self-inspection task satisfies:

[0104] (12);

[0105] Wherein, is the total risk cost, is the total number of cycles of the inspection period, are the inspection durations corresponding to each inspection period respectively.

[0106] More generally, the above formula (12) can be expressed as:

[0107] (13).

[0108] For DDQN, the reward obtained at is the discounted accumulation of the rewards obtained from the current moment to the end of the entire scenario. The reward function in the embodiments of the present invention should aim to minimize the total risk cost as the goal.

[0109] Exemplarily, the input state is input into the inspection planning model. The inspection planning model will consider the target personnel that can be captured within the coverage range covered by the target camera (determined by the personnel position information. If there is a target personnel who has moved far away from the target camera, this target personnel will not be selected as the inspection target because the distance is too far, and the target camera cannot accurately track, and even if focused, the imaging is not clear), and consider the construction risk coefficient of the target personnel and other factors, and output an inspection probability between 0 and 1 for each target personnel = .

[0110] In the embodiments of the present invention, when it comes to target personnel with different construction risk coefficients, the overall risk cost needs to be considered. For example, the risk coefficients of operation points such as high-altitude operation and live operation are relatively high, while the risk coefficient of ground non-live operation is relatively low. At this time, combined with the position of the target camera and the construction risk coefficient, the optimal path is sought. Especially when the positions of the on-site personnel move, because the position and field of view range of the target camera are recorded in the state matrix, according to the properties of the Markov chain, the deployed inspection planning model can directly calculate the optimal strategy in combination with the current on-site state, without the need to recalculate all paths, meeting the real-time requirements of the control system, especially more meaningful for complex environmental states.

[0111] Specifically, in steps S3 to S4, after obtaining the patrol probabilities of all target personnel, the target personnel corresponding to the maximum value is taken as the patrol target, and the focusing direction of the target camera is controlled to be aligned with the patrol target to complete the shooting of the patrol image until all patrol cycles are completed.

[0112] Exemplarily, the safety management platform is deployed in the cloud platform, and various types of violation detection models can be deployed thereon, such as helmetless identification, fence-crossing identification, etc. The safety management platform receives the patrol images uploaded in step S4 for violation identification operations. If a violation operation occurs for the patrol target, a prompt message is sent, such as issuing an instruction to trigger an audible and visual alarm.

[0113] Compared with the prior art, the patrol method for the construction environment disclosed by the present invention has the following beneficial effects:

[0114] 1. Aiming at the problem in the prior art that the camera autofocus operation depends on safety supervisors, resulting in high skill requirements for supervisors and limited real-time performance, the present invention monitors construction personnel through high-precision image extraction technology, thereby providing a more reliable guarantee for construction safety. Through automated and intelligent image processing technology, the present invention not only reduces the work burden of safety supervisors, but also enhances the real-time performance and accuracy of construction site safety management.

[0115] 2. By obtaining the construction risk coefficients of target personnel, combining personnel location information and camera information, and using the patrol planning model to output patrol probabilities, it is possible to accurately select the personnel with the highest risk from among many unpatrolled personnel as the patrol target, making the patrol resources most effectively utilized, avoiding blind patrols, and improving the pertinence and efficiency of patrols.

[0116] 3. Considering personnel location and camera information helps to plan a more reasonable patrol path, reduce time waste and distance consumption during the patrol process, improve the overall patrol efficiency, and shorten the time required to complete all patrol cycles.

[0117] 4. Focusing on patrolling personnel with high construction risk coefficients can timely detect potential safety hazards and violation behaviors during their construction process, and timely take measures to correct and handle them, reducing the occurrence probability of construction accidents, and ensuring the life safety of construction personnel and the smooth progress of construction projects.

[0118] 5. Based on multi-source data such as construction risk coefficients, personnel location information, and camera information, through analysis and decision-making by the patrol planning model, intelligent patrol management is realized. This data-driven method can formulate patrol strategies more objectively and accurately, avoiding interference and subjectivity of human factors.

[0119] See Figure 6 ,Figure 6 It is the structural frame of an inspection system 100 for a construction environment provided by an embodiment of the present invention. The inspection system 100 for the construction environment includes:

[0120] An information acquisition module 11, configured to acquire the construction risk coefficient and personnel location information of all target personnel in the construction environment in each inspection cycle, and acquire the camera information of the target cameras in the construction environment;

[0121] An inspection probability generation module 12, configured to input the construction risk coefficient, the personnel location information, and the camera information into an inspection planning model, so that the inspection planning model outputs the inspection probability of each target personnel;

[0122] An inspection target determination module 13, configured to select the target personnel corresponding to the maximum value from all inspection probabilities as the inspection target for the current inspection cycle;

[0123] An inspection control module 14, configured to control the focusing direction of the target camera to be aligned with the inspection target in the current inspection cycle until all inspection cycles complete the inspection.

[0124] Specifically, the construction risk coefficient is calculated according to the risk degree value corresponding to the personnel information among the target personnel; wherein, the personnel information includes at least one of basic information, team information, and construction information, and each personnel information corresponds to a risk degree value.

[0125] Specifically, the basic information includes attribute information and first historical violation information; when the personnel information includes basic information, the determination method of the first risk degree value corresponding to the basic information includes:

[0126] Dividing the historical data into positive and negative samples; wherein, the positive samples include data records without safety accidents, and the negative samples include data records with safety accidents;

[0127] For the attribute information in the negative samples, performing clustering analysis using a clustering algorithm to obtain several attribute categories;

[0128] Calculating the contribution rate of each attribute category for safety accidents, and integrating the contribution rates of all attribute categories to obtain a contribution rate set;

[0129] Obtaining the target contribution rate matching the target personnel from the contribution rate set, and obtaining the violation risk coefficient corresponding to the first historical violation information;

[0130] Using the sum of the target contribution rate and the violation risk coefficient as the first risk degree value.

[0131] Specifically, calculating the contribution rate of safety accidents occurring in each attribute category includes:

[0132] For each attribute category, count the number of positive samples belonging to the current attribute category in the positive samples, and count the number of negative samples belonging to the current attribute category in the negative samples;

[0133] Determine the contribution rate of safety accidents occurring in the current attribute category based on the number of positive samples and the number of negative samples.

[0134] Specifically, the first historical violation information includes a first time interval from the violation time of the target person to the current time, and the violation risk coefficient is inversely proportional to the first time interval.

[0135] Specifically, the team information includes second historical violation information and historical working hours. The second historical violation information is a second time interval from the violation time of any person in the team to the current time; when the personnel information includes team information, the method for determining the second risk degree value corresponding to the team information includes:

[0136] Calculate a first risk coefficient based on the second time interval, and calculate a second risk coefficient based on the historical working hours; wherein, the second time interval and the first risk coefficient are inversely proportional, and the historical working hours and the second risk coefficient are inversely proportional;

[0137] Use the sum of the first risk coefficient and the second risk coefficient as the second risk degree value.

[0138] Specifically, the construction information includes the construction type and its corresponding construction risk level; when the personnel information includes construction information, the third risk degree value corresponding to the construction information is directly proportional to the construction risk level.

[0139] Specifically, the patrol planning model is a reinforcement learning model, and a reward function is set in the patrol planning model. The patrol planning model takes the maximum value output by the reward function as the optimization goal; wherein, the reward function is related to the construction risk coefficient.

[0140] It should be noted that the working processes of each module in the patrol system 100 of the construction environment described in the embodiments of the present invention can refer to the working process of the above-mentioned patrol method for the construction environment, and will not be elaborated here.

[0141] Furthermore, an embodiment of the present invention also provides a computer-readable storage medium storing a computer program, wherein when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the patrol method for the construction environment described in any of the above embodiments.

[0142] The implementation of all or part of the processes in the above-described method embodiments of the present invention can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed, the steps of the above-described method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0143] The above is the preferred implementation manner of the present invention. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements are also regarded as the protection scope of the present invention.

Claims

1. A method for inspecting a construction environment, characterized in that: include: In each inspection cycle, the construction risk factor and personnel location information of all target personnel in the construction environment are obtained, as well as the camera information of the target cameras in the construction environment; Inputting the construction risk coefficient, the personnel location information and the camera information into the inspection planning model, so that the inspection planning model outputs the inspection probability of each target personnel; Select the target person corresponding to the maximum value from all inspection probabilities as the inspection target of this inspection cycle; In this inspection cycle, the focus direction of the target camera is controlled to be aligned with the inspection target until all inspection cycles are completed; The construction risk coefficient is calculated based on the risk level value corresponding to the personnel information of the target personnel; the personnel information includes at least one of basic information, team information and construction information, and each personnel information corresponds to a risk level value; The basic information includes attribute information and first historical violation information; when the personnel information includes basic information, a method for determining a first risk level value corresponding to the basic information includes: dividing historical data into positive and negative samples; wherein the positive samples include data records in which no safety accidents have occurred, and the negative samples include data records in which safety accidents have occurred; for the attribute information in the negative samples, clustering analysis is performed using a clustering algorithm to obtain a number of attribute categories; the contribution rate of each attribute category to safety accidents is calculated, and the contribution rates of all attribute categories are integrated to obtain a contribution rate set; a target contribution rate matching the target personnel is obtained from the contribution rate set, and a violation risk coefficient corresponding to the first historical violation information is obtained; the sum of the target contribution rate and the violation risk coefficient is taken as the first risk level value; The calculation of the contribution rate of each attribute category to safety accidents includes: for each attribute category, counting the number of positive samples belonging to the current attribute category in the positive samples, and counting the number of negative samples belonging to the current attribute category in the negative samples; determining the contribution rate of the current attribute category to safety accidents based on the number of positive samples and the number of negative samples.

2. The inspection method for a construction environment as claimed in claim 1, characterized in that: The first historical violation information includes a first time interval between the target person's violation time and the current time, and the violation risk coefficient is inversely proportional to the first time interval.

3. The inspection method for a construction environment as claimed in claim 1, characterized in that: The team information includes second historical violation information and historical working hours, wherein the second historical violation information is a second time interval between the violation time of any person in the team and the current time; When the personnel information includes team information, the method for determining the second risk level value corresponding to the team information includes: Calculating a first risk coefficient according to the second time interval, and calculating a second risk coefficient according to the historical working hours; wherein the second time interval and the first risk coefficient are inversely proportional, and the historical working hours and the second risk coefficient are inversely proportional; The sum of the first risk coefficient and the second risk coefficient is taken as the second risk degree value.

4. The inspection method for a construction environment as claimed in claim 1, characterized in that: The construction information includes the construction type and its corresponding construction risk level; when the personnel information includes construction information, the third risk level value corresponding to the construction information is in direct proportion to the construction risk level.

5. The inspection method for a construction environment as claimed in claim 1, characterized in that: The inspection planning model is a reinforcement learning model, in which a reward function is provided. The inspection planning model takes the maximum value of the reward function output as an optimization target; wherein the reward function is related to the construction risk coefficient.

6. A construction environment inspection system, characterized in that: include: An information acquisition module is used to obtain the construction risk factor and personnel location information of all target personnel in the construction environment in each inspection cycle, and to obtain the camera information of the target camera in the construction environment; An inspection probability generation module, used for inputting the construction risk coefficient, the personnel location information and the camera information into the inspection planning model, so that the inspection planning model outputs the inspection probability of each target personnel; Inspection target determination module, used to select the target personnel corresponding to the maximum value from all inspection probabilities as the inspection target of this inspection cycle; The inspection control module is used to control the focus direction of the target camera to align with the inspection target in this inspection cycle until all inspection cycles are completed; The construction risk coefficient is calculated based on the risk level value corresponding to the personnel information of the target personnel; the personnel information includes at least one of basic information, team information and construction information, and each personnel information corresponds to a risk level value; The basic information includes attribute information and first historical violation information; when the personnel information includes basic information, a method for determining a first risk level value corresponding to the basic information includes: dividing historical data into positive and negative samples; wherein the positive samples include data records in which no safety accidents have occurred, and the negative samples include data records in which safety accidents have occurred; for the attribute information in the negative samples, clustering analysis is performed using a clustering algorithm to obtain a number of attribute categories; the contribution rate of each attribute category to safety accidents is calculated, and the contribution rates of all attribute categories are integrated to obtain a contribution rate set; a target contribution rate matching the target personnel is obtained from the contribution rate set, and a violation risk coefficient corresponding to the first historical violation information is obtained; the sum of the target contribution rate and the violation risk coefficient is taken as the first risk level value; The calculation of the contribution rate of each attribute category to safety accidents includes: for each attribute category, counting the number of positive samples belonging to the current attribute category in the positive samples, and counting the number of negative samples belonging to the current attribute category in the negative samples; determining the contribution rate of the current attribute category to safety accidents based on the number of positive samples and the number of negative samples.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the inspection method for the construction environment as described in any one of claims 1 to 5.

Images