Early warning method and system for abnormal posture of operating personnel in limited space

Abstract

The invention relates to an early warning method and system for abnormal attitude of an operator in a limited space, and the method comprises the steps: obtaining image data of the operator in the limited space through binocular vision equipment, and obtaining the position coordinates of each key node; obtaining a space coordinate matrix and a relation matrix of the nodes according to the position coordinates of the key nodes and the connection relation between the nodes; calculating the variable quantity of each node relative to all nodes at a certain time interval, and constructing an attitude change matrix of each node in combination with the space coordinate matrix and the relation matrix of the nodes; and classifying the key nodes to obtain a classification cluster to which each node belongs, and by calculating the distance between each node and the center vector of the classification cluster to which the node belongs and the variable quantity of the center vector of each classification cluster, judging whether the posture of the operator is abnormal or not and performing early warning when the posture is abnormal. The method aims at the time-space relationship of all nodes of a to-be-detected individual target during syncope or falling, and carries out the precise recognition and discrimination of the abnormal personnel through the coordination of the time-space node relationship.

Description

Technical Field

[0001] The present application relates to the technical field of worker detection, and more specifically, to a method and system for early warning of abnormal posture of workers in confined spaces. Background Art

[0002] Existing workers entering deep foundation pits or underground pipe galleries of power transmission lines mainly rely on voice response mechanisms to confirm the workers' activity status. In addition, the space of deep foundation pits or underground pipe galleries of power transmission lines is small, complex, and difficult to move freely, which has certain limitations in the recognition of people's postures.

[0003] At present, the confined space posture recognition technology includes the use of a variety of sensors, such as cameras, lidar, inertial measurement units (IMUs), etc., to recognize and track the postures of objects or human bodies in confined spaces. However, the deployment of a large number of sensors on the bodies of operators seriously affects operation and maintenance activities and poses certain safety risks.

[0004] Prior art document 1 (CN109919134B) proposes a vision-based method for detecting abnormal behavior of operating vehicle personnel. The method first collects image information of the driver and the people around him in the vehicle in all directions; then, extracts 2D images, and uses the deep learning algorithm in the openpose database to annotate the skeletal nodes of all people in the image; then, combined with the vehicle driving status, defines the interfering person and the interfered person, and calculates the dangerous distance threshold of each key limb part of the interfered person. Subsequently, the spatial distance between the wrist node of the interfering person and the key limb part of the interfered person is calculated based on the binocular ranging source; finally, the abnormal behavior of the interfering person is detected by judging the time when the spatial distance is within the dangerous distance threshold. However, the detection of abnormal personnel in the prior art requires the use of multiple objects for judgment, such as detecting the distance between the interfering person and the interfered person through binocular vision, and judging abnormal behavior through the distance threshold. The detection method is complex and involves multiple objects. It is not suitable for the detection of a single object, and the detection speed is slow and the accuracy is not high. Summary of the invention

[0005] In order to solve the deficiencies in the prior art, the present invention provides a method for early warning of abnormal posture of personnel working in confined spaces, which can solve the technical problems of slow recognition speed and insufficient accuracy in the prior art for detecting personnel in confined spaces.

[0006] The present invention adopts the following technical solution.

[0007] A method for early warning of abnormal posture of workers in confined spaces, comprising the following steps:

[0008] Obtain image data of workers in confined spaces through binocular vision equipment, set key nodes of workers and obtain the position coordinates of each key node in combination with image data;

[0009] According to the position coordinates of the key nodes and the connection relationship between the nodes, the spatial coordinate matrix and relationship matrix of the nodes are obtained;

[0010] Calculate the change of each node relative to all nodes at a certain time interval, and construct the posture change matrix of each node by combining the spatial coordinate matrix and relationship matrix of the node;

[0011] The key nodes are classified to obtain the classification cluster to which each node belongs. By calculating the distance between each node and the central vector of the classification cluster to which it belongs and the change of the central vector of each classification cluster, it is judged whether the posture of the operator is abnormal.

[0012] Preferably, obtaining the position coordinates of the key nodes of the operator's body in the confined space and obtaining the depth data specifically includes:

[0013] Use binocular vision equipment to obtain depth images of operators and construct a three-dimensional coordinate system based on the acquired images;

[0014] Calculate the depth data of each pixel. In the depth image, each pixel corresponds to a depth value, which represents the distance from the point to the camera.

[0015] Obtain the position information of the key nodes of the human body in the image, including the position coordinates and direction, and obtain the position of the key nodes of the human body in three-dimensional space;

[0016] The obtained position information of key nodes of the human body is used to monitor and track the posture of the operator.

[0017] Preferably, a total of n key nodes are set, and the obtained position coordinates are the relationship coordinates between the node position information and time t. The position coordinates p of key node i are i (t)Satisfy:

[0018] p i (t) = (x i (t), y i (t), z i (t))

[0019] Among them, x i (t), y i (t), z i (t) respectively represent the three-dimensional space coordinates of the i-th key node at the t-th time, i∈[1,n].

[0020] Preferably, the node relationship matrix and the relationship matrix between the key nodes are constructed as follows:

[0021] According to the position coordinates p of the key node i i (t), and get the spatial coordinate matrix of n key nodes at the tth moment:

[0022]

[0023] Among them, x i (t), y i (t), z i (t) represents the three-dimensional space coordinates of the i-th key node at the t-th time;

[0024] Set the connection relationship between the i-th key node and the j-th key node to w ij , when two key nodes are strongly connected w ij The value is 1, when there is no connection between two key nodes ij The value of is 0, and the relationship matrix W of all key nodes is obtained:

[0025]

[0026] Preferably, the calculation time interval is Δt, and the change of each key node relative to all key nodes is as follows:

[0027] Calculate the change of node i relative to all key nodes within Δt time, and obtain the change matrix g of key node i relative to all key nodes i :

[0028]

[0029] Where, Δg ij It represents the change of key node i relative to key node j when the time interval is Δt, and i, j∈[1,n].

[0030] Preferably, the posture change matrix is ​​formed based on the change matrix corresponding to each key node, wherein the key node relationship matrix G corresponding to the i-th key node is i satisfy:

[0031]

[0032] Among them, W is the relationship matrix of all key nodes.

[0033] Preferably, the key nodes are classified to obtain the classification cluster to which each node belongs, specifically including:

[0034] Use the clustering algorithm to divide the relationship changes of each key node into K clusters, and calculate the center vector of the kth cluster as u k :

[0035]

[0036] Among them, c k represents the number of key nodes in the kth cluster.

[0037] G i represents the pose change matrix of the i-th key node in the K-th cluster.

[0038] Preferably, calculating the distance between each node and the center vector of the classification cluster to which it belongs and the change amount of the center vector of each classification cluster specifically includes:

[0039] Calculate the distance d between key node i and the center vector of cluster k to which it belongs ik :

[0040] d ik =||G i -u k ||2

[0041] Among them, || ||2 represents the L2 normal form;

[0042] The change in the depth value of each cluster center vector Δu ij The calculation formula is:

[0043] Δu ij =||u i -u j ||2

[0044] In the formula, u i 、u j They represent the center vectors of the i-th and j-th clusters respectively.

[0045] Preferably, determining whether the posture of the operator changes suddenly specifically includes:

[0046] The distance d between the calculated key node i and the center vector of the cluster k to which it belongs is respectively ik Deviation from the preset threshold ∈ i Compare the change Δu of the depth value of each cluster center vector ij and the preset difference threshold ω i The size of the operator's posture is determined based on the comparison result to determine whether there is a sudden change in the operator's posture. The judgment criteria are as follows:

[0047]

[0048] If the distance d between the key node i and the center vector of the cluster k to which it belongs ik Greater than the preset deviation threshold ∈ i , or the change in the depth value of each cluster center vector Δu ijGreater than the preset difference threshold ω i , it is judged that the operator's posture has changed suddenly, which means that the operator's posture is abnormal and an alarm is issued. Otherwise, it is judged that the operator's posture has not changed suddenly, which means that the operator is normal.

[0049] The present invention also provides a confined space operator posture abnormality warning system using the confined space operator posture abnormality warning method, comprising: a data acquisition module, a position processing module, a node processing module, a node classification module, and an abnormality judgment module;

[0050] The data acquisition module includes a binocular vision device for collecting images of workers in confined spaces;

[0051] The position processing module obtains the position information of the key nodes of the human body of the operator in the confined space according to the collected image processing, and obtains the depth data;

[0052] The node processing module constructs the node relationship matrix between key nodes and the posture change matrix of all nodes according to the position information of key nodes of the human body;

[0053] The node classification module is used to classify key nodes to obtain the classification cluster to which each node belongs;

[0054] The abnormality judgment module determines whether the operator's posture is abnormal by calculating the distance between each node and the central vector of the classification cluster to which it belongs and the change of the central vector of each classification cluster, and issues an alarm when it is abnormal.

[0055] The present invention also provides a terminal, including a processor and a storage medium;

[0056] The storage medium is used to store instructions;

[0057] The processor is used to operate according to the instructions to execute the steps of the method for early warning of abnormal posture of workers in confined spaces.

[0058] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method for early warning of abnormal posture of workers in confined spaces are implemented.

[0059] The beneficial effect of the present invention is that, compared with the prior art, the present invention proposes a method for quickly identifying abnormal postures of workers in confined spaces and issuing early warnings, using binocular vision to monitor the postures and depth data of workers, obtaining the position information of key nodes of the human body, and then calculating the relative motion amplitude between each key node, and judging whether the worker has abnormal posture through a threshold; the present invention makes judgments from two plane angles and three-dimensional depth data, and identifies and distinguishes through the spatiotemporal relationship of all nodes of the individual target to be detected when fainting or falling, and through the coordination of this spatiotemporal node relationship, without the need for precise identification of the posture of the person, providing fast, real-time and accurate abnormal identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of the method for early warning of abnormal posture of workers in confined spaces in the present invention;

[0061] Figure 2 It is a structural schematic diagram of the abnormal posture warning system for workers in confined spaces in the present invention. DETAILED DESCRIPTION

[0062] In order to make the purpose, technical scheme and advantages of the present invention clearer, the technical scheme of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

[0063] like Figure 1 As shown, the present application proposes a method for warning abnormal posture of workers in confined spaces, which uses binocular vision to monitor the working state of the human body, obtains the position information of the key nodes of the human body, and then calculates the relative motion amplitude between each key node, and judges whether the worker has abnormal posture through the threshold. The abnormal detection object in the present invention is the abnormal posture of a single person in a confined space, which is used to judge abnormal behaviors such as fainting caused by toxic gases or lack of oxygen in the confined space, and the time-space relationship of the key nodes of the workers when they faint or fall, and the coordination of this time-space node relationship is used to judge abnormalities.

[0064] The method specifically comprises the following steps:

[0065] Step 1, obtaining image data of key nodes of the human body of the operator in the confined space through a binocular vision device, and obtaining the position coordinates of each key node according to the image data;

[0066] Compared with traditional monocular vision or inertial measurement unit (IMU) and other technologies, binocular vision can provide higher-precision posture recognition and tracking. Binocular vision can obtain depth data, making posture recognition more accurate.

[0067] Binocular vision can cover a larger spatial range and is suitable for scenes of different scales. By setting different camera layouts and parameters, it can adapt to scene requirements of different scales.

[0068] Since binocular vision can capture object information from multiple angles and perspectives, it can reduce the impact of factors such as occlusion and light changes and improve the robustness of gesture recognition.

[0069] Compared with the gesture recognition method based on depth sensor, binocular vision has higher real-time performance. Modern binocular vision technology can achieve high frame rate image acquisition and processing to achieve real-time gesture recognition and tracking.

[0070] Binocular vision does not require contact with the object being measured and can perform posture monitoring without affecting human operations, thereby better ensuring the safety of workers.

[0071] In summary, binocular vision posture recognition technology has the advantages of high precision, large range, strong robustness, strong real-time and non-contact, so it has a positive effect in monitoring the working status of human body in confined space and ensuring safety. The present invention adopts binocular vision equipment to collect images of workers in confined space.

[0072] Binocular vision equipment includes two cameras, which can simultaneously shoot objects to obtain the depth information of objects. In confined spaces, binocular vision is used to monitor the working status of human bodies and obtain the position information of key nodes of the human body, including monitoring the working status of human bodies in confined spaces through binocular vision, obtaining the position coordinates of key nodes of the human body, using the image in machine vision as the coordinate system, and obtaining the depth data. The depth data is the distance between each key node and the binocular vision module to obtain the coordinates of each key node. The specific process is as follows:

[0073] Acquire an image of the target work object through a binocular camera, and align the two images captured by the binocular camera to determine their relative position and direction;

[0074] Then use the disparity method or triangulation method to calculate the depth data of each pixel. In the depth image, each pixel corresponds to a depth value, which represents the distance from the point to the camera.

[0075] Select the key nodes of the target work object, formulate the acquired position information through the coordinate system of the image in the machine vision, and obtain the position of the key nodes of the human body in three-dimensional space. Use the acquired position information of the key nodes of the human body to monitor and track the posture of the human body.

[0076] The three-dimensional position coordinates of each key node are obtained by combining the key nodes and the three-dimensional coordinate system; specifically, the three-dimensional position coordinates of the key nodes are obtained by taking the binocular camera as the coordinate origin and combining the positions of the key nodes in the depth image.

[0077] Assume that the number of key nodes of the human body is n, and the key nodes set include at least arms, legs, torso, and head. The position of each node can be regarded as a function of time. The position may be different at different times. The obtained position coordinates are expressed as the relationship between the position information and time t. For the i-th key node, its position coordinate p i (t)Satisfy:

[0078] p i (t) = (x i (t), y i (t), z i (t))

[0079] Among them, x i (t), y i (t), z i (t) respectively represent the three-dimensional space coordinates of the i-th node at the t-th time, i∈[1,n].

[0080] Step 2, according to the position coordinates of the key nodes of the human body and the connection relationship between the nodes, the spatial coordinate matrix and the relationship matrix of the nodes are obtained;

[0081] According to the position coordinates of each key node obtained in step 1, the spatial coordinate matrix P(t) of n key nodes at time t is obtained as follows:

[0082]

[0083] Furthermore, a node relationship matrix between key nodes is constructed to obtain the interactions and dynamic changes between nodes;

[0084] The relationship between the key nodes of the human body in the confined space can be divided into the connection relationship between the arms, legs, torso, and head. Specifically, the connection relationship between the i-th node and the j-th node is set as w ij , 0≤w ij ≤1, when the two nodes are strongly connected, its value is 1, when the two nodes have no connection, it is 0, and the node relationship matrix W is obtained:

[0085]

[0086] Step 3: Calculate the change of a node relative to other nodes at a time interval of Δt, and construct the posture change matrix G of all nodes g ;

[0087] Calculate the change of node i relative to all nodes within Δt time, and get the change matrix g of node i relative to all nodes i :

[0088]

[0089] Where, Δg ij represents the change of node i relative to node j at time interval Δt, and i, j∈[1,n];

[0090] Combine the change matrices corresponding to each node to form a posture change matrix, where the node relationship matrix G corresponding to the i-th node is i :

[0091]

[0092] Where W is the node relationship matrix obtained in step 2.

[0093] Step 4: Classify the key nodes in combination with the node relationship matrix, and determine whether the operator's posture is abnormal by calculating the distance between each node and the center vector of the classification cluster to which it belongs and the change in the center vector of each classification cluster.

[0094] Step 4-1: Use the clustering algorithm to divide the relationship changes of each node into K clusters, and calculate the center vector of the kth cluster as u k :

[0095]

[0096] Among them, c k represents the number of nodes in the kth cluster;

[0097] G i represents the posture change matrix of the i-th node in the K-th cluster.

[0098] Step 4-2, calculate the distance between each node and the center vector of the cluster to which it belongs, and compare it with the preset deviation threshold ∈ i The comparison is made to find out the degree of deviation of the node from the cluster to which it belongs.

[0099] By calculating the distance between node i and the center vector of its cluster k, and comparing this distance with the preset deviation threshold, the deviation degree of the node relative to its cluster is obtained as follows:

[0100] The distance d between node i and the center vector of cluster k to which it belongs ik :

[0101] d ik =||G i -u k ||2

[0102] Here, || ||2 denotes the L2 form.

[0103] Furthermore, the distance d between the calculated node i and the center vector of the cluster k to which it belongs is ik With deviation threshold ∈ i Compare and get the degree of deviation of the node relative to its cluster:

[0104] If d is satisfied ik ≤∈ i , it means that the node has not deviated from its cluster, otherwise it means that the node has deviated from its cluster, and it is judged that the operator's posture has a sudden change;

[0105] Step 4-3, calculate the change in the depth value of each cluster center vector, measure the posture difference between different clusters, and compare it with the preset difference threshold ω i The comparison is performed, and the comparison result and the degree of deviation of the node relative to its cluster are combined to determine whether the operator's posture has undergone a sudden change.

[0106] The change in the depth value of each cluster center vector Δu ij The calculation formula is:

[0107] Δu uj =||u i -u j ||2

[0108] In the formula, u i 、u j They represent the center vectors of the i-th and j-th clusters respectively.

[0109] If the node does not deviate from its cluster, the change in the depth value of each cluster center vector Δu is further determined. ij and the preset difference threshold ω i The size of Δu ij ≤ω i , it means that the operator's posture has not changed suddenly, otherwise it is judged that the operator's posture has changed suddenly;

[0110] Therefore, according to the relationship between the distance between each node and the center vector of the cluster to which it belongs and the threshold, and the relationship between the change in the depth value of the center vector of each cluster and the threshold, it is judged whether the posture of the operator has changed suddenly. The judgment criteria are as follows:

[0111]

[0112] That is, if the distance or depth value between any node and the center vector of the cluster to which it belongs exceeds the threshold, it means that the posture has undergone a sudden change. At this time, it means that the operator may have an abnormal accident and needs to issue an alarm and provide emergency rescue. If the posture does not undergo a sudden change, it means that the operator is normal.

[0113] The judgment criteria consider the distance between each node and its cluster and the change of the center vector between different clusters. By adjusting the threshold, it can flexibly adapt to different scenarios and needs.

[0114] When judging anomalies, when the distance between a node and its cluster or the change in the center vector between clusters exceeds a threshold, the system will issue an alarm and may trigger further emergency measures.

[0115] like Figure 2 As shown, the present invention also proposes a confined space operator posture abnormality warning system, and the above-mentioned confined space operator posture abnormality warning method can be implemented based on the system, and the system includes: a data acquisition module, a position processing module, a node processing module, a node classification module, and an abnormality judgment module;

[0116] Among them, the data acquisition module includes a binocular vision device for collecting images of workers in confined spaces;

[0117] The position processing module obtains the position information of the key nodes of the human body of the operator in the confined space according to the collected image processing, and obtains the depth data;

[0118] The node processing module constructs the node relationship matrix between key nodes and the posture change matrix of all nodes according to the position information of key nodes of the human body;

[0119] The node classification module is used to classify key nodes to obtain the classification cluster to which each node belongs;

[0120] The abnormality judgment module determines whether the operator's posture is abnormal by calculating the distance between each node and the central vector of the classification cluster to which it belongs and the change of the central vector of each classification cluster.

[0121] The beneficial effect of the present invention lies in that, compared with the prior art, the method proposed in the present invention makes judgments from two plane angles and three-dimensional depth data, and identifies and distinguishes through the spatiotemporal relationship of all nodes of the individual target to be detected when fainting or falling, and through the coordination of this spatiotemporal node relationship, thereby providing fast, real-time and accurate abnormality recognition without the need for precise identification of the person's posture.

[0122] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0123] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive lists) of computer readable storage medium include: portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), static random access memories (SRAM), portable compact disk read-only memories (CD-ROM), digital versatile disks (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or convex structures in grooves on which instructions are stored, and any suitable combination thereof. Computer readable storage medium used here is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by waveguides or other transmission media (e.g., light pulses by optical fiber cables), or electrical signals transmitted by wires.

[0124] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0125] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages-such as Smalltalk, C++, etc., and conventional procedural programming languages-such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network-including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA) may be personalized by utilizing the state information of a computer-readable program instruction, and the electronic circuit may execute a computer-readable program instruction, thereby realizing various aspects of the present disclosure.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for early warning of abnormal posture of workers in confined spaces, characterized in that: The steps include: Obtain image data of workers in confined spaces through binocular vision equipment, set key nodes of workers and obtain the position coordinates of each key node in combination with image data; According to the position coordinates of the key nodes and the connection relationship between the nodes, the spatial coordinate matrix and relationship matrix of the nodes are obtained; Calculate the change of each node relative to all nodes at a certain time interval, and construct the posture change matrix of each node by combining the spatial coordinate matrix and relationship matrix of the node; The key nodes are classified to obtain the classification cluster to which each node belongs. By calculating the distance between each node and the center vector of the classification cluster to which it belongs and the change of the center vector of each classification cluster, it is determined whether the posture of the operator is abnormal and an early warning is issued in case of abnormality.

2. The method for early warning of abnormal posture of workers in confined spaces according to claim 1, characterized in that: Obtain the position coordinates of the key nodes of the operator's body in the confined space and obtain the depth data, including: Use binocular vision equipment to obtain depth images of operators and construct a three-dimensional coordinate system based on the acquired images; Calculate the depth data of each pixel. In the depth image, each pixel corresponds to a depth value, which represents the distance from the point to the camera. Obtain the position information of the key nodes of the human body in the image, including the position coordinates and direction, and obtain the position of the key nodes of the human body in three-dimensional space; The obtained position information of key nodes of the human body is used to monitor and track the posture of the operator.

3. The method for early warning of abnormal posture of workers in confined spaces according to claim 2, characterized in that: Assume that there are n key nodes in total, and the position coordinates obtained are the relationship coordinates between the node position information and time t. The position coordinates of key node i are p i (t)Satisfy: p i (t)=(x i (t),y i (t),z i (t)) Among them, x i (t), y i (t), z i (t) respectively represent the three-dimensional space coordinates of the i-th key node at the t-th time, i∈[1,n].

4. The method for early warning of abnormal posture of workers in confined spaces according to claim 3 is characterized in that: The node relationship matrix and relationship matrix between key nodes are constructed as follows: According to the position coordinates p of the key node i i (t), and get the spatial coordinate matrix of n key nodes at the tth moment: Among them, x i (t), y i (t), z i (t) represents the three-dimensional space coordinates of the i-th key node at the t-th time; Set the connection relationship between the i-th key node and the j-th key node to w ij , when two key nodes are strongly connected uj The value is 1, when there is no connection between two key nodes uj The value of is 0, and the relationship matrix W of all key nodes is obtained:

5. The method for early warning of abnormal posture of workers in confined spaces according to claim 1, characterized in that: The calculation time interval is Δt, and the change of each key node relative to all key nodes is as follows: Calculate the change of node i relative to all key nodes within Δt time, and obtain the change matrix g of key node i relative to all key nodes i : Where, Δg ij It represents the change of key node i relative to key node j when the time interval is Δt, and i, j∈[1,n].

6. The method for early warning of abnormal posture of workers in confined spaces according to claim 5, characterized in that: The posture change matrix is ​​constructed based on the change matrix corresponding to each key node, where the key node relationship matrix G corresponding to the i-th key node is i satisfy: Among them, W is the relationship matrix of all key nodes.

7. The method for early warning of abnormal posture of workers in confined spaces according to claim 6, characterized in that: The key nodes are classified to obtain the classification clusters to which each node belongs, including: Use the clustering algorithm to divide the relationship changes of each key node into K clusters, and calculate the center vector of the kth cluster as u k : Among them, c k represents the number of key nodes in the kth cluster. G i represents the pose change matrix of the i-th key node in the K-th cluster.

8. The method for early warning of abnormal posture of workers in confined spaces according to claim 7, characterized in that: Calculating the distance between each node and the center vector of the classification cluster to which it belongs and the change in the center vector of each classification cluster specifically includes: Calculate the distance d between key node i and the center vector of cluster k to which it belongs ik : d ik =||G i -u k ||2 Among them, || ||2 represents the L2 normal form; The change in the depth value of each cluster center vector Δu ij The calculation formula is: Δu uj =||u i -u j ||2 In the formula, u i 、u j They represent the center vectors of the i-th and j-th clusters respectively.

9. The method for early warning of abnormal posture of workers in confined spaces according to claim 8, characterized in that: Determining whether the operator's posture is abnormal specifically includes: The distance d between the calculated key node i and the center vector of the cluster k to which it belongs is respectively ik Deviation from the preset threshold ∈ i Compare the change Δu of the depth value of each cluster center vector ij and the preset difference threshold ω i The size of the operator's posture is determined based on the comparison result to determine whether there is a sudden change in the operator's posture. The judgment criteria are as follows: If the distance d between the key node i and the center vector of the cluster k to which it belongs ik Greater than the preset deviation threshold ∈ i , or the change in the depth value of each cluster center vector Δu ij Greater than the preset difference threshold ω i , it is judged that the operator's posture has changed suddenly, which means that the operator's posture is abnormal and an alarm is issued. Otherwise, it is judged that the operator's posture has not changed suddenly, which means that the operator is normal.

10. A confined space operator posture abnormality warning system using the confined space operator posture abnormality warning method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, location processing module, node processing module, node classification module, abnormality judgment module; The data acquisition module includes a binocular vision device for collecting images of workers in confined spaces; The position processing module obtains the position information of the key nodes of the human body of the operator in the confined space according to the collected image processing, and obtains the depth data; The node processing module constructs the node relationship matrix between key nodes and the posture change matrix of all nodes according to the position information of key nodes of the human body; The node classification module is used to classify key nodes to obtain the classification cluster to which each node belongs; The abnormality judgment module determines whether the operator's posture is abnormal by calculating the distance between each node and the central vector of the classification cluster to which it belongs and the change of the central vector of each classification cluster, and issues an alarm when it is abnormal.

11. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 9 are implemented.

Images